Cryptopolitan

Berkshire Hathaway’s stock (NYSE: BRK.A, BRK.B) is rallying today because the company still trails the wider market, trades below last year’s richer valuation, holds almost $400 billion in cash, and may have bought back billions of dollars of its own shares.
Barron’s reported that the rally carried the conglomerate to an eight-month high this week. The gain also came while technology shares fell, drawing money toward a large defensive company with businesses in insurance, rail, energy, manufacturing, retail, and finance.
The Class B shares closed Tuesday at $512.37, their best finish since November 28, when they ended at $513.81. They had closed Friday at $511.54, or 5.2% below the record close of $539.80 set on May 2, 2025.
That record came one day before Warren Buffett said he would leave the CEO job at the end of 2025. The Class A shares ended Tuesday at $768,010, also their highest close since November 28’s $770,100. Friday’s Class A close of $766,600 was 5.3% under its record of $809,350.
Berkshire Hathaway attracts buyers as tech falls and UBS raises its target
Both Berkshire share classes are up only about 1% in 2026, compared with a return of roughly 9% for the S&P 500. The company has also trailed stocks connected to two of its largest operations. Union Pacific (NYSE: UNP), the closest listed comparison for BNSF, has gained about 30% this year. Chubb (NYSE: CB), a major property and casualty insurer, has also posted a much larger gain.
UBS Group (NYSE: UBS) analyst Brian Meredith raised his Class A price target by 3%, taking it to $877,848 from $854,596, while keeping a Buy rating. That target is about 15% above the share price used in his report. Brian estimated Berkshire’s intrinsic value at nearly $800,000 per Class A share, around 5% above the current price.
Brian also raised his profit forecasts. His 2026 estimate for the Class B shares increased 1.3% to $21.05, while his 2027 figure rose 0.8% to $21.32. He cited “modestly higher earnings at BNSF and lower catastrophe losses” during the second quarter. The disaster-loss figure relates to Berkshire’s large property and casualty insurance business.
Berkshire Hathaway (NYSE: BRK.A, BRK.B) now trades at about 1.4 times an estimated book value of roughly $535,000 per Class A share. The earlier estimate for the end of the second quarter was about $522,000. The current ratio sits near the lower end of its range in recent years and below the 1.8 times recorded in May 2025, when Class A traded near $810,000.
The company is due to publish its second-quarter results within the next two weeks. That report will include shareholder equity and an updated book-value figure. Berkshire’s cash balance is also helping the stock while traders cut technology exposure. The Technology Select Sector SPDR Fund (NYSE Arca: XLK) fell another 2% Tuesday, and Berkshire has often traded against the tech sector’s direction this year.
Rail, insurance, buybacks, and stock holdings add support to Berkshire Hathaway
Brian said Berkshire appeared to repurchase about $8.5 billion of its stock during the second quarter, based on an ownership filing from Warren. He called the possible purchase a “bullish sign.” It would rank among the company’s largest quarterly buybacks. Barron’s placed the likely total between $5 billion and $11 billion after reviewing the filing earlier this month.
There is no peer comparison on the open market for Berkshire Hathaway since the stock is valued at around $1.1 trillion and consists of a diversified portfolio of businesses. However, there are some stocks related to Berkshire that have performed much better than the latter. The railroad company CSX (NASDAQ: CSX), for instance, is up by more than 50% this year.
Berkshire also controls the world’s largest property and casualty insurance capital base. Its insurance companies include auto insurer Geico and reinsurer Gen Re. Listed insurers have posted strong gains, with Chubb and Everest Group (NYSE: EG) rising about 15% to 20%.
The listed stock portfolio has added more value during the quarter and across 2026. CNBC’s portfolio tracker puts its size near $360 billion. Apple (NASDAQ: AAPL), Berkshire’s largest holding, reached a new high Tuesday near $340 and is up about 25% this year. The position is worth roughly $77 billion.
Coca-Cola (NYSE: KO) also reached a record on Tuesday after reporting second-quarter earnings. Its shares rose about 6% that day, traded near $89, and brought their yearly gain close to 29%. Berkshire owns more than $35 billion of the stock. Bank of America (NYSE: BAC) reached a 52-week high on Monday and has gained about 10% this year. Berkshire’s stake is valued at above $30 billion.
The smartest crypto minds already read our newsletter. Want in? Join them.

The Vice President of Bitcoin Strategy at Strive, Joe Burnett, described the past week as “possibly one of the worst weeks in the history of Bitcoin.”
The statement suggested that large holders ought to trust institutional custodians more than hardware wallets.
Why did Strive’s VP of strategy call out the worst week in Bitcoin’s history?
A firmware exploit led to the loss of ~1,082 BTC, worth about $70 million, from Coldcard hardware wallets.
The theft caused Burnett to warn large holders about hardware wallets, encouraging institutional custodians instead. The thief pounced on weak seed randomness before carting away users’ funds.
The flaw was found in version 4.0.0. Coinkite shipped that version from a code commit on March 1, 2021, and patched it in version 4.21. Coinkite is the company behind Coldcard.
The price of Bitcoin stayed more or less the same, hovering around $63,000 through the weekend. Although Santiment recorded the most negative Bitcoin social sentiment ever, a reaction that could be traced to the horror of watched-over keys failing.
Why Burnett says the setup was more critical than the mistake
What caused Burnett to be uneasy wasn’t carelessness per se. According to Burnett, victims of the attack purchased a genuine device, generating their seed offline and sticking to the recommended playbook, but still lost their coins. He surmised that a self-custody arrangement with a single point of failure is not strong enough and can break in different ways.
Burnett differentiated between the risk of a self-custody arrangement and how institutional custodians operate. In his view, large firms like Fidelity and BitGo operate differently from the exchange minefield of the pre-COVID era.
Burnett bluntly stated that institutional custodians look to avoid the same trap.
CZ, the founder of Binance, responded to the same attack by urging holders to spread their coins across wallets as “nothing is 100%.”
Strive’s position in the Bitcoin trade
Strive is a company built around holding Bitcoin. Strive trades on the Nasdaq with the ASST ticker, with Matt Cole as its CEO. The company ranks as the seventh-largest public company holder with 20,000 BTC valued at nearly $1.3 billion as of late July.
Burnett is one of Bitcoin’s loudest proponents. He predicts Bitcoin will be at $11 million per Bitcoin by Q1 of 2036. He posits that AI-driven deflation will force central banks to keep expanding the money supply, which will lead to such prices.
If you're reading this, you’re already ahead. Stay there with our newsletter.

After first-half trading turnover fell 54.6% from a year earlier, South Korea’s five won-based crypto exchanges are rebuilding their businesses around bank and brokerage partnerships, institutional services and regulatory cleanup.
Due to trading volume being down by 55%, exchanges are changing their business models to survive.
What is happening to South Korea’s crypto market?
The combined first-half turnover at Upbit, Bithumb, Coinone, Korbit and Gopax reached $366.58 billion, representing a 54.6% drop year over year. The five companies reached a cumulative volume of roughly 17.3 trillion won from July 1 to 27. This represents a 16.9% drop from the same stretch a month earlier.
The market has been shrinking against local stocks for months. In July, the five exchanges averaged 597.8 billion won ($406.5 million) in daily turnover, equal to just 1.59% of the KOSPI stock market’s daily average. That ratio was above 11% in January before falling to around 6% in March and April and about 2% in May.
For comparison, domestic crypto volume reportedly hit as high as 21 trillion won in a single day in November 2024, when traders bet on crypto-friendly policy from a second Trump administration.
Notably, Bitcoin’s average daily move ran 1.25% in the first half and altcoins 1.79%, both below the KOSPI’s 4.67%, with Bitcoin stuck between $60,000 and $63,000. Some retail investors have shifted funds to the stock market to chase the rising KOSPI index.
Korbit sold 15 BTC and 60 ETH over ten days to raise roughly $11.6 million (1.6 billion KRW) in cash. Dunamu (operator of Upbit) saw its first-quarter revenue and operating profit decline by 55% and 78% year-over-year, respectively.
A director at Tiger Research referred to the current situation as “second crypto winter,” considering the sharp contraction in total crypto market cap and the fact that no new projects are emerging.
However, the winter lifted for 48 hours when the KOSPI shed as much as 864 trillion won in market value on July 28 and 29. The five exchanges traded $964.11 million on July 28, up 82.5% from the prior month’s daily average, as a circuit breaker halted a stock drop of more than 10%. Upbit’s USDT turnover hit 200.045 billion won on July 29, roughly double its Bitcoin volume.
Why are exchanges turning to banks and brokerages?
Due to the thinner volume, liquidity was pushed towards the largest exchanges. Upbit’s July turnover fell by 10% to about 11.7 trillion won, yet its market share climbed from 62.3% to 67.4%. Meanwhile, Bithumb dropped by 26.6% to 4.7 trillion won and saw its share slip to 27.1%, leaving a 40.3-percentage-point gap between the two.
Upbit and Bithumb’s combined share was reportedly above 90% in early July, and a Kaiko report on the Korean market, written by Presto Labs, found the top two handling nearly 96% of domestic volume historically.
Upbit capitalized on its advantage by waiving fees on six stablecoins from July 26, briefly lifting its USDT turnover to nearly 160 billion won, roughly four times Bithumb’s.
Coinone is attempting to expand beyond token trading by adding a stock-trading tab to its app Home Screen that routes users to Korea Investment & Securities’ web platform, the first product since the brokerage took a stake in May. Korea Investment & Securities and OKX Ventures each paid about 80 billion won for roughly 20% holding, a 160-billion-won deal that left founder Cha Myung-hoon’s side with 30.4%.
Mirae Asset Consulting bought 97.15% of Korbit, a deal the Fair Trade Commission cleared partly because Korbit’s market share sits near 0.5%. Korbit plans to relaunch as DigitalX and expand into stablecoins, real-world assets, tokenized securities and custody, aiming at corporate and institutional clients rather than retail traders.
The move follows eight straight years of operating losses since 2018.
Did every exchange find easy solutions?
Unlike Korbit, Bithumb held talks with Kiwoom Securities, Kakao, Toss and Meritz Securities about cooperation or investment, but none could satisfy Bithumb’s terms regarding price, control and structure.
Any potential deal is automatically complicated by the tangled ownership web between Bithumb Holdings, Bidet and T Scientific, unresolved items including an FIU administrative lawsuit, a Bitcoin misdelivery sanction and a pending VASP renewal.
Gopax, the smallest at about 0.1% share, put recovery ahead of growth. Streami named former AWS Korea executive Kim Na-young as chief and prioritized resolving GoFi, a suspended deposit product with losses approaching 100 billion won after FTX collapsed in 2022.
Binance paid out about $70 million through September 2023, but nothing has been paid since. Regulators insist the repayment must be made in full before Gopax can receive its VASP renewal. The exchange raised its won deposit rate from 1.30% to 1.50%, and it’s still under Upbit’s 2.1% and Bithumb’s 2.2%.
The smartest crypto minds already read our newsletter. Want in? Join them.

Investor Chamath Palihapitiya has published a map of where he expects money to move through the AI market on X, telling his followers that the fastest cash sits in power and data-center real estate.
However, he also stated that the durable margins will belong to “harnesses” and the applications built on top of them.
Where did Palihapitiya say the cash will land first in his guide?
The Social Capital founder tagged his post on X as his “AI investing guide” as of August 2026.
He categorized areas for investments into layers, and the first layer he treated was what he calls LPS, short for land, power, and shell. It refers to the physical footprint of a data center before any chips go in.
Palihapitiya wrote that it is “still the most obvious and fastest path to cash on cash returns.” He added, “Lots of value can be assembled and traded quickly at this layer. And as data centers get more pushback, energized land can explode in value. Very bullish here.”
Palihapitiya said he and his partner, Anita Verma-Lallian, have locked in close to 6GW of power running through 2029.
He had previously stated that zoning-approved land and silicon access hand their owners negotiating leverage over everyone downstream.
Why does Palihapitiya believe harnesses and applications will be the winners?
Above the concrete and the power lines, Palihapitiya’s pick is the harness. He stated in a post made in July, “A modern harness + open model will crush your token consumption but keep your performance.”
A harness is the software wrapped around an AI model that decides what the model sees, which tools it can call, and when it stops, according to a Hugging Face glossary published on May 25.
Anthropic’s Claude Code, OpenAI’s Codex, and Google’s Antigravity are all harnesses. Claude Code is referred to as “the agentic harness around Claude” in its documentation.
Palihapitiya stated that “the harness helps enterprises owns their proprietary context (what Alex Karp calls their ‘alpha’),” which to him includes their data, workflows, and business rules, among others.
His thesis also stretches to applications, as he says they will be another long-term winner. He wrote, “Every company, with the right harness, can now imbue their alpha into the software that runs their company.”
Are people agreeing with Palihapitiya?
Some industry figures have chipped in their takes on Palihapitiya’s post, with many supporting it, especially his point on harnesses.
Xiaoyin Qu, the founder of Tycoon AI, expressed more support for harnesses, stating that a harness “will create margin regardless of if the model gets commoditized,” because the right one can unlock large, long-horizon jobs that are worth more than any single model output.
Aaron Levie, Box’s CEO, in response to a different post that highlighted the performances of various AI agents, stated that the harness is “going to become the most important variable” in the AI stack, sitting right next to raw model capability.
In mid-July, Palihapitiya made a post on X that questioned the current state of AI spending, asking if it was paying off for anyone beyond the handful of firms already collecting the money, pointing to buyers who can now spend $0.50 per million leading-edge tokens instead of $56 for the same volume.
It may seem that the harness call is his answer to his own complaint because if models get cheap and interchangeable, the money moves to whoever controls the data, the workflows, and eventually the applications sitting on top.
If you're reading this, you’re already ahead. Stay there with our newsletter.

Social media chatter about leading cryptocurrency, Bitcoin, has plunged to the most negative depths on record in the immediate aftermath of the Coldcard hardware wallets exploit that drained an estimated 1,082 BTC, worth roughly $70 million, from affected users.
The social sentiment readings tracked by Santiment confirmed the sour mood after a security event that affected almost 1,200 self-custody wallets.
Why the Coldcard wallet hack was a sucker punch
Crypto holders store their tokens on hardware wallets to safeguard them against the various operational and security risks that target centralized exchanges every day. In fact, the “Not your keys, not your crypto” phrase became popular after users lost access to their tokens after exchanges failed in the past.
In essence, your crypto should be safe as long as your keys are securely stored. However, all of that logic failed in the Coldcard incident, which Santiment believes was enough to compound the historic breakdown in sentiment.
In the first full day of trading since the Coldcard exploit, Santiment data logged positive Bitcoin commentary at its lowest level compared to negative posts. The Saturday reading is the worst the firm has recorded since it adopted its current tracking method across X, Reddit, Telegram and other social media platforms.
Bitcoin sentiment reading for July 31. Source: Santiment
Block traced the cause to a firmware build error present since Coinkite, the firm behind Coldcard, shipped version 4.0.0 from a code commit dated March 1, 2021. Coinkite patched the issue in release 4.21.
Did the Coldcard exploit affect Bitcoin price?
Ironically, Bitcoin price has held steady near $63,000 as of this Saturday report. The leading cryptocurrency ended July in the green even though it stepped back by about 1.3% over the last 24 hours, per CoinMarketCap data.
Bitcoin maintaining above the $60,000 level represents a win for market watchers through a period where victims saw estimated losses climb from $38 million to $70 million.
Cryptopolitan first reported the losses around 594 BTC (about $38 million) around the time the hack first happened. Those numbers went up sharply throughout the day, as Galaxy Research’s most recent estimate came in as $70 million lost from 1,196 addresses.
What CZ and the market are watching next
Binance founder Changpeng Zhao weighed in on X, writing that “even hardware wallets can have bugs” and suggesting holders spread funds across several wallets, while adding that the approach carries its own risks and “nothing is 100%.” His post drew nearly 2,000 likes within hours.
For the market, the near-term question is the attacker’s consolidated stash. A large transfer toward an exchange could add short-term supply pressure. Traders are also eyeing whether Bitcoin holds $60,000; a sustained break could open the way toward $58,000, while a recovery of $64,000 would ease the immediate downside.
If you're reading this, you’re already ahead. Stay there with our newsletter.

Fintech prime broker Clear Street has launched a platform that enables accredited investors to acquire pre-IPO shares in late-stage tech companies. The initiative allows investors to invest in an investment area that heretofore was reserved for only large institutions. Databricks, an AI and data software company with a valuation of $188 billion, is the first business to be introduced on this platform.
The platform is emblematic of a larger trend in AI investment. With the likes of Databricks, Anthropic, and OpenAI staying private longer than earlier tech behemoths, an increasing share of the value of the industry is being created before public debuts. Clear Street is betting that making this investing approach accessible will provide additional funds, although it is unclear whether this will spur AI innovation or lead only to higher private valuations.
Margin loans on private stakes set the platform apart
Chief Executive Officer and co-founder of Clear Street, Uri Cohen, said that the objective of the firm is “to remove friction and give more people the ability to invest in more products.” He claimed that a major part of making today’s wealth takes place in private markets and that demand from retail investors keeps growing.
In contrast to the majority of private market platforms, Clear Street intends to include financing as well. The company aims to provide loans based on pre-IPO investments by taking care of asset servicing and risk on its own, which is uncommon, given that it’s difficult to sell private stock before the actual exit.
Clear Street hopes to onboard 30 startups by the end of the year. The company is focusing on tech startups worth $5-$20 billion that are about six months to two years away from going public. The company is also launching a specialized private company research unit led by Owen Lau.
Databricks anchors the launch at a $188 billion valuation
Databricks is a natural choice for a first public offering. Its Lakehouse Platform melds together all the benefits of data lakes and data warehouses, while being equipped with the likes of Delta Lake and MLflow for enterprise AI development. Its valuation of $188 billion shows a strong interest by investors in the infrastructure that underpins AI.
However, Databricks is still far from being publicly traded. As reported by Kiplinger, referencing information from June 18, the company is expected to have its IPO in 2027, while Anthropic will go public in 2026. The extended timeline gives investors only a few regulatory means through which they can invest before the company goes public, providing an avenue for firms like Clear Street.
Wealth is pooling in private markets before the bell rings
Clear Street is not the only entity making strides to fulfill such a need. Goldman Sachs has also broadened the reach of its private company investment offerings to wealthy customers, a move that reflects a bigger trend in the industry as early-stage companies put off going public and create more value without entering public markets.
The IPO market has shown hopeful first indications of recovery. According to data from Renaissance Capital, total IPO filings for the year as of August 1 reached 155, which is an increase of 10.7% compared to the same period last year. The number of IPOs offered this year is 93, with total proceeds of $144.0 billion, thanks to the successful IPO by SpaceX worth $75 billion.
Meanwhile, Clear Street is delaying its own entry. The startup that was worth about $12 billion this year has moved its potential IPO to 2027 as it has achieved cash flow-positive operations and completed a $ 400 million investment-grade bond issue.
The valuation-versus-revenue gap hangs over the bet
The valuation of Databricks, which stands at $188 billion, raises another, more general question surrounding private AI markets: to what extent is today’s valuation driven by expectations for the future and by current performance of the company?
One thing that has been pointed out by Cryptopolitan earlier is that AI companies’ valuations look overly high compared to their actual revenues, thus indicating that investors make bets on future returns rather than invest in the companies based on their current performance.
Clear Street offers a platform that enables investors to gain early access to this trading opportunity. In the event that AI companies deliver the anticipated growth, investing prior to IPOs may yield significant rewards. However, if revenue takes longer to catch up with valuations, early investors may be taking on increased risk.
The smartest crypto minds already read our newsletter. Want in? Join them.

The Hong Kong law enforcement agencies reportedly received 25 reports of investment scams during the last week of July. The case involved romance and online dating scams duping almost HK$70 million (around US$9 million) in losses.
TRM Labs data shows that the money involved in the illicit activities went up to US$158 billion in 2025. This was a surge of around 145% year-on-year. Investment scams account for a large share of such Financial crimes. It has made it difficult for exchanges, regulators, and crypto firms to regain public trust.
A HK$26 million loss that started with insurance advice
One of the fresh cases mentioned that a woman in her fifties who worked in insurance was introduced to a lady who allegedly wanted to discuss insurance matters with her. However, she was later introduced to a man claiming to be in the automobile business, helping someone with insurance inquiries. Their WhatsApp conversations gradually became romantic.
The man said that he was an experienced investor and convinced her to use a crypto trading application. He then introduced her to a fake platform manager to manage her wallet. After half a year, she gave over 4 million HKD (approx worth $510k) in cash and transferred almost 22 million HKD (approx worth $2.8 million) to the accounts given by the scammers. Her account seemed to yield profits of over 800%.
When she tried to withdraw her money, her request was denied, leaving her with a loss of more than 26 million HKD (approx worth $3.32 million).
The pattern is a well-known one in the scamming world. Back in May 2026, the South China Morning Post reported another case of a woman from Hong Kong losing over HKD 1 million after being lured by a manipulated advertisement on Facebook that referred her to an AI-based investment site made popular on WhatsApp. In one week alone, authorities received over 70 investment scam complaints, claiming losses of over HKD 50 million.
Scams evolve from romantic entanglements to AI-promoted investments. However, the basic principles of the scam have changed very little. It begins by earning the trust of would-be marks and showing phony gains before forbidding them to take their money out.
Why investment fraud now dominates Hong Kong’s loss column
Hong Kong’s crime data shows why investment fraud has become a priority. Deception accounted for 48.5% of all reported crime in 2025. Although total scam cases fell 2.9% to 43,212, online investment fraud rose 30.7% to 5,135 cases, while losses jumped 58.4% to HK$3.58 billion, according to the Anti-Deception Coordination Centre (ADCC).
Investment scams represented only 11.9% of deception cases but generated 44.1% of all financial losses. Average losses also increased from roughly HK$580,000 to HK$700,000 per case.
In October 2025, the HK police formed a Virtual Asset Intelligence Taskforce in order to enhance collaboration with Customs and financial authorities and improve methods of prevention of money laundering involving cryptocurrencies. The ADCC had also directed more focus on the campaigns linked to awareness-raising in investment fraud in 2026.
Data suggests that authorities are more concerned with the economic consequences of cases than with their sheer numbers. In spite of the fact that shopping and employment scams exist more than cryptocurrency investment scams, the damage resulting from these scams has shown that, in fact, the latter is worse. Additionally, the probability of reclaiming the lost funds drops drastically once the funds have been turned into cryptocurrency and moved between wallets or foreign exchanges, emphasizing the need for prevention, rapid reporting, and cross-border cooperation.
Cambodia and Myanmar remain scam hubs
Although the victims are based in Hong Kong, the perpetrators seldom happen to be local. As per the UN Office on Drugs and Crime report, it is found that online scams in 2025 cost victims from East Asia, Southeast Asia, Australia, and New Zealand between $88.3 billion and $114.1 billion, which is three times more than the situation in 2023. The report showed that Cambodia and Myanmar were among the main areas for criminal groups that run romance and crypto scams.
Many employees within those firms are also victims themselves. The International Organization for Migration estimates that there are at least 300,000 people involved in these fraud operations in South-East Asia as a result of false job advertising.
The cases demonstrate that contemporary investment fraud combines social media, chat platforms, and traditional bank systems, resulting in cross-border fraud networks. As oversight of licensed cryptocurrency firms increases, criminals make use of the weak point of the system: human trust in the technology of blockchain.
It is for this reason that the most recent cases in Hong Kong have importance beyond their immediate locale. They show that successful adoption of cryptocurrencies depends not only on implementing regulation, but on breaking up the globally operating scam networks that continue to prey on investors.
If you're reading this, you’re already ahead. Stay there with our newsletter.

OpenAI has reportedly terminated a series of ChatGPT accounts that are linked to a Cambodian criminal organization that used the chatbot to commit fraud in cryptocurrency investment. It presents another case where AI and crypto have been misused by organized criminals.
This comes in when digital assets-linked scams are on the rise, which is now urging exchanges, token developers, and artificial intelligence companies to take action to prevent losses to victims. “Pig-butchering” scams are already widespread in Southeast Asia, where fraudsters swindle crypto payments with either dating or investment schemes while using AI-generated content.
How the operators wired ChatGPT into the con
According to OpenAI’s report, the investigation began after receiving a tip-off from WhatsApp, and they have since traced the activity back to accounts alleged to be based in Poipet, a Cambodian border city already known to international investigators for its connection to large-scale scams.
According to the firm’s statement, the operators exploited ChatGPT by creating fake identities, translating talks into different languages, and composing messages to win people’s confidence. Furthermore, the chatbot produced marketing materials for the fake investment schemes.
At the heart of the plot was cryptocurrency. Victims were lured into love or friendship conversations, then redirected to fake cryptocurrency and spot-gold trading platforms. To further add to the illusion, the perpetrators employed ChatGPT to fabricate fake passports, legal notices, stock purchase confirmations, and phony crypto trading dashboards.
The AI firm indicated that the fraud occurred in three stages, which allowed the criminals to trick victims into sending them cryptocurrency first and then request more money through some supposed activation fees, taxes, or regulatory fines.
The trafficking layer sitting behind the accounts
OpenAI said its investigation also uncovered evidence of labor trafficking.
Some companies have taken advantage of ChatGPT to design ads for recruiting “chatters” in Poipet. The ads include attractive offerings like free tickets, lodging, food, and help with visas. Other discussions have to do with making spreadsheets to keep track of employees’ debts, money taken from their salaries, fines for inappropriate behavior, immigration status, detention, or attempted escape.
The results bear a resemblance to the study conducted by Elliptic, whereby most of the actors behind romance and investment scams happen to be victims of human trafficking made to work in organized crime operations throughout the Southeast Asian region.
Before shutting down the accounts, OpenAI provided information from the investigation to industry stakeholders and law enforcement. The company surmises, based on internal communication, that the network has reached out to hundreds of victims and that their losses may total thousands of dollars.
This case illustrates a larger trend regarding cybersecurity. As organized crime syndicates can utilize generative artificial intelligence and cryptocurrency payment systems, preventing criminals from using advanced AI tools might become as key as blocking illegal wallets or shutting down money laundering operations.
Illicit crypto flows hit a record just as the ban dropped
Given the scale of crypto crime, OpenAI’s move is huge. According to TRM Labs’ Crypto Crime Report released on January 28, illicit cryptocurrency flows were estimated to amount to $158 billion in 2025. This was an increase of nearly 145% from the previous year and is the largest figure in five years. While the share of illicit activity within on-chain volume dropped from 1.3% in 2024 to 1.2% in 2025. This indicates that crypto has been adopted more widely.
The report also indicates that stablecoins are responsible for around 84% of verified fraud money inflow, confirming their being the number one method of payment for criminals engaged in organized scams. However, OpenAI’s shutdown is aimed at a single operation done within a criminal infrastructure.
Why AI companies have become a chokepoint for scam networks
OpenAI’s efforts are huge because generative AI can already carry out the most labor-intensive part of scams – convincing victims to trust scammers.
A study by researchers from Amrita Vishwa Vidyapeetham, Ca’ Foscari University of Venice, University of Melbourne, and Ben-Gurion University of the Negev indicated how an AI-enabled Claude agent managed to outperform a human scammer in the process of making a pig-butchering scam. It revealed that 46% of the participants used the software suggested by the AI after one week of messages, while only 18% of people used it with the human operator. Moreover, the participants concluded that the AI was much more reliable.
The researchers conducted interviews with 145 people who were previously involved in scamming operations, including survivors of trafficking operations in different facilities in Cambodia, Myanmar, and Laos. According to the findings, one of the functions of using AI in scam operations is its ability to automate the trust-building stage in the scams, and blocking access to sophisticated models can lead to the disruption of the operation process even before the victims have been influenced to give up their money.
The smartest crypto minds already read our newsletter. Want in? Join them.

Bitcoin managed to finish July up nearly 8 percent. BTC defied rate-hike fears, the AI-driven sell-off and a major exchange security breach. However, traders are now waiting on cues from the jobs data and a potential return of spot Bitcoin exchange-traded funds. The OG crypto price is still down by more than 28% year to date (YTD).
Bitcoin price dropped by almost 3% on the last day of July 2026. The drawdown was insufficient to derail the rally. Analysts attributed the lack of leveraged exposure on the derivatives markets after a selloff in late June, which saw the price fall below $58,000 and trigger large liquidations of leveraged positions. This, they argued, made the markets less vulnerable to further declines as the Federal Reserve began its rate-hike cycle.
Did Bitcoin avoid a deeper selloff?
According to the report, the average daily liquidations since early July remained well below the peak of $400 million to $000 million recorded at various points through the year. “Crypto fell less than levered equity themes because the forced-selling fuel was already spent,” the analysts wrote.
It was also reflected in the performance of crypto compared to the broader markets. On Friday, the final day of July, both Bitcoin price and Ether fell sharply despite South Korea’s Kospi index rising over 15 percent and U.S. equity futures hitting record highs.
BTC remained resilient despite the technical development, where one of the leading self-custody hardware wallet providers, Coldcard, became the victim of a major security breach. Earlier, it was reported that the attack had resulted in the theft of at least $38 million in Bitcoin. Later, Galaxy Research reportedly claimed that the losses were around $70 million in crypto assets.
Bitcoin Monthly Returns; Source: CoinGlass
Coinglass data shows that Bitcoin had printed a red index and posted a dump of more than 14% in the second quarter of 2026. However, the third quarter is still green, giving hope of a rally ahead.
How much longer can the rally last?
While the immediate outlook remained broadly positive, several analysts were cautious, noting that markets were now entering a period of heightened volatility due to conflicting views about the prospects for rate hikes. An analyst noted that investors were caught in a tug of war between dovish and hawkish scenarios, which was weighing on high-beta assets. Markets are entering a new volatility regime as we transition from a period of policy tightening to speculation about cuts. Simultaneously, there are concerns about higher rates dampening risk appetite.
Bitfinex analysts expect the same dynamics to play out in August, with positioning remaining net long, albeit tentatively, while a rate hike by the Federal Reserve continues to be priced in. “The signal for traders which has not yet fired is the institutional buying of bitcoin at aggressive levels,” they wrote, noting that the open interest for Bitcoin futures had remained relatively static at around 750,000 BTC throughout the month. This reflected the reluctance of traders to build leveraged positions despite the bullish onboarding of new investors.
The base case for August is range-bound Bitcoin with limited short-term volatility if yields remain range-bound or ETF flows turn consistently positive. Meanwhile, a neutral outcome from the Federal Reserve was acceptable, but persistent dollar strength, higher yields and weak ETF inflows would weigh on prices.
Diminished prospects for corporate holders
The cautious outlook for the markets in the months ahead has implications for corporate players. As the largest institutional bitcoin holder with 843,775 BTC as of late July, Strategy has been heavily impacted by the recent price action, recording a second consecutive quarter of losses.
According to the firm’s earnings report released on July 30, it recorded a net unrealized loss of $8.22 billion, driven by an $8.32 billion loss in the second quarter. “We acquired additional bitcoin and treasury securities and reduced our convertible debt issuance during the quarter while navigating muted bitcoin prices and market conditions,” said Michael Saylor, chairman of Strategy, in a statement.
For now, the focus for corporate holders appears to be on demand generation. “The critical test for institutional demand will come in the form of increased inflows into spot bitcoin ETFs,” noted Bitfinex analysts, who suggested that their appearance in early July had failed to deliver on initial expectations.

Don’t just read crypto news. Understand it. Subscribe to our newsletter. It's free.

The trading platform trade.xyz has started to repay traders whose leveraged trades were cancelled after a single wrong SK hynix order in South Korea, which later led to around $60 million in long liquidations on Hyperliquid. This situation shows how one mistake in the traditional stock market can easily carry over to the world of on-chain derivatives, especially now that tokenized-stock perpetuals are becoming a bigger part of crypto trading.
The payouts, confirmed by trade.xyz on Aug. 1, resolve a pricing incident that unfolded on July 27. The market at the center of the event, a USDC-margined SK Hynix perpetual, was far from niche. According to Galaxy Research, it was the largest builder-deployed market on Hyperliquid, with $638 million in open interest as of July 30.
How one share in Seoul reached an onchain order book
The chain of events started in South Korea’s conventional stock market. Galaxy Research expert Will Owens stated that when the pre-market time slot of NextTrade started at 8:00 in the morning on July 28, one share of SK hynix was worth 1,272,000 won (around $868), which is around 30% less than the previous price. Although this price met the daily limit imposed by the stock exchange, buyers entered the market, and the price started to rise to almost 1.7 million won within a couple of minutes.
By then, the reaction had already reached the cryptocurrency markets.
The xyz: SKHYNIX perpetual product of trade.xyz tracks the dollar value of one share of SK hynix and switches to outside pricing once NextTrade opens. The company stated in its incident report on July 29 that the $868 transaction was valid, having been confirmed by different independent market data providers, thus demonstrating a proper operation of the oracle.
Why the liquidations hit crypto traders so hard
trade.xyz calculates its mark price by considering the average of three sources of data and introduces a smoothing method to eliminate sudden fluctuations. According to Owens, the service neutralized about 11% of the almost 30% incorrect movement. Still, there was much more left that caused liquidation.
The mark price itself dropped about 18.7% almost immediately, while the open interest fell within minutes from $481 million to $331 million.
According to Galaxy Research, approximately 960 accounts were liquidated with long positions amounting from $57 million to $80 million. As prices dropped, profitable short positions were automatically deleveraged, and 406 liquidated long positions were taken up by a backstop address until its liquidation.
Following the mentioned events, trade.xyx announced that the mark price had decreased from $1,127.9 down to $917.25 by 23:01 UTC on July 27, which coincided with the opening of NextTrade’s pre-trade session.
What the distributions actually cover
The refunds were calculated by trade.xyz using the reference rate of $1,115.5. As such, traders whose losses did not exceed $10,000 were entitled to automatic refunds, while those whose losses were larger had to apply for them.
The company makes it clear that refunds are a “one-time discretionary decision” and not a regular obligation, since everything went right with the system.
In addition, the company mentioned that it is speeding up efforts to improve its pricing strategy in tough market conditions where, in the matter of price, the orders of the company are prioritized over trades made by other market players.
The stakes for Hyperliquid’s builder-market boom
The situation also emphasizes the significance of builder markets within Hyperliquid’s ecosystem.
trade.xyz, a perpetual futures branch of tokenization protocol Unit, was the first market to open under Hyperliquid’s HIP-3 enhancement, which was rolled out in mainnet in October 2025. Under the upgrade, any team that stakes 500,000 HYPE (around $27 million) can create its own perpetual futures markets.
trade.xyz has made up over 90% of HIP-3 open interest and almost 98% of builder market trades, as per Galaxy Research.
The market has also grown exponentially. From a mere 2% earlier in the year, HIP-3 now accounts for almost half of all Hyperliquid’s perpetual trading transactions. The volume on builder marketplaces alone is around $98 billion for the last thirty days, while HIP-3’s open interest has reached approximately $3.6 billion.
The incident illustrates that a pricing glitch of a single stock-linked perpetual is no longer an isolated case. It can influence a large part of the entire onchain derivatives market.
The timing added to the perplexity of the case. SK hynix faced significant selling pressure that week, sliding down by 14.65% to reach 1.55 million won. It further declined after the announcement of its quarterly results the next day. The crashes triggered circuit breakers for several trading days for the first time in the history of the Korean market, according to Galaxy Research, thus giving credence to the initial decline prior to the identification of the erroneous transaction.


Don’t just read crypto news. Understand it. Subscribe to our newsletter. It's free.

The XRP Ledger (XRPL) is preparing for its next major software upgrade, with RippleX Head of Product Jazzi Cooper revealing five protocol upgrades that would give the network much more power on tokenization, institutional finance and real-world asset (RWA) adoption.
The xrpld 3.3.0 release is expected to be released next week. As with all XRP Ledger protocol changes, however, the new features will only be activated once all independent validators are approved by the network via an amendment voting process.
Announcing the new features on X, Cooper said that the XRP Ledger has already demonstrated its ability to support tokenized assets at scale and is expanding to support global transfers, trading, collateralization, and settlement.
The five amendments include Confidential MPT, Batch, Permission Delegation, Sponsored Fees and Reserves, and Dynamic MPT, which will improve the usability of the network for financial institutions and enterprise applications.
The upgrade is coming at a time when there is increasing competition between blockchain networks to attract institutional tokenization projects. Financial firms are in need of blockchain technology to tokenize traditional assets (like government bonds, real estate, equities, and private credit) for payment and to reduce costs.
Confidential MPT brings privacy to tokenized assets
One of the most interesting additions is Confidential MPT (Multi-Purpose Tokens) which makes privacy features possible with elliptic curve cryptography and zero-knowledge proofs.
The amendment allows token issuers and holders to keep the balance and transaction amount private but allow for entities such as regulators or auditors to access it when needed.
The feature addresses one of the biggest barriers preventing financial institutions from adopting public blockchain infrastructure, where transaction data is typically visible to everyone
The proposed Batch amendment provides for the ability to group multiple transactions involving different accounts into a single atomic operation.
This enables complex financial workflows, including delivery-versus-payment (DvP) settlements, to run entirely or fail together, reducing settlement risk and enhancing efficiency for institutional trading and tokenized asset transfers.
Permission Delegation improves security
Another amendment, Permission Delegation, allows for organizations to delegate very narrowly defined transaction permissions without giving up control of their primary signing keys.
The feature is designed for treasury teams and financial institutions that need to authorize routine transactions, the report says. At the same time, it allows them to maintain strict control over critical issuance and reserve accounts.
Further, the Sponsored Fees and Reserves proposal aims to simplify onboarding by allowing banks, issuers or platforms to pay XRP transaction fees and reserve requirements on behalf of their users.
This means new users can use applications on XRPL without having to buy XRP in order to sign up for or pay transaction fees – a longstanding usability obstacle for newcomers. Users would still have access to their wallets and private keys.
Dynamic MPT brings flexibility after issuance
The fifth amendment, Dynamic MPT, would allow the issuers to modify the select token properties on XRP Ledger. These include features such as transfer fees, token metadata, other predefined features after token issuance, without the need for a fresh token issuance.
Copper said: “The release is currently anticipated for next week. As always, these amendments will only activate following validator approval.” She also urged validator operators to go through the amendments which will be available as they come.
The proposed release comes just days after the XRP Ledger activated the fixCleanup3.2.0 amendment, which brought several bug fixes to vaults, lending protocol, permissioned domains, Multi-Purpose Tokens and permissioned decentralized exchange.
More than 80% of validators supported the amendment before it was activated, demonstrating the continuing involvement in the governance of the network.
Like previous upgrades, xrpld 3.3.0 will first make the proposed amendments available for voting. They will only become part of the XRP Ledger protocol after validators reach consensus, ensuring that the decentralized network, not Ripple, ultimately decides whether new features are activated.
As with all XRPL protocol upgrades, none of the proposed amendments are applied immediately after the release of the software. Validators must first approve each amendment using the decentralized governance of the network.
Once an amendment has at least 80% approval from trusted validators for two consecutive weeks, it is enabled on the XRP Ledger. This governance model is intended to make sure that new features are widely supported before they are permanently introduced.
Don’t just read crypto news. Understand it. Subscribe to our newsletter. It's free.

Arthur Hayes disposed of 2,364.38 ETH for 4.3 million USDC last Friday, according to tracking platform Lookonchain, resulting in a possible loss of $241,000. The sale happened as the price of crypto, in general, was on the decline with the whole cryptocurrency market falling by about 2 percent within the past 24 hours.
Arthur Hayes(@CryptoHayes) bought high and sold low again!
Over the past 2 hours, he deposited 2,364.38 $ETH into Cumberland and Galaxy Digital, receiving 4.3M $USDC in return.
His selling price was $1,821, resulting in a loss of $241K (-5.3%).
He had previously bought 7,213… pic.twitter.com/4AVZpjANZD
— Lookonchain (@lookonchain) August 1, 2026
The transaction did not get attention for its size. What made this transaction unique is that the former BitMEX CEO and closely monitored macro investor Hayes has been a prominent supporter of Ethereum for a long time. His choice to liquidate some of his investments at a loss has sparked new discussions about how investors adapt to the downturn in the current market.
Selling into a market already in the red
Hayes sold the ETH at an average price of about $1,821 per coin, according to Lookonchain. The tokens were transferred to trading firms Cumberland and Galaxy Digital in exchange for USDC over roughly two hours.
The transaction happened on a day of poor performance for digital currencies. Market capitalization in cryptocurrency fell by approximately 2.1% to $2.25 trillion, while Bitcoin was trading around $63,000, a 2.7% decrease. Ether fell approximately 3.1% down to $1,860 with CryptoQuant estimating the quoted price to be $1,864.85, a 3.56% drop from its value a day before.
Despite the fact that Hayes’s sale was only a small amount of Ethereum’s daily trading volume, it got critical attention as it converted paper losses into realized losses at a time of weak investor sentiment.
A losing streak that started in July
The sale marks the end of a downtrend that started several weeks prior. Based on Lookonchain data, Hayes collected 7,213 ETH between July 15 and July 28 for about $13.87 million, with an average cost of approximately $1,923 per ETH. However, even before Friday’s trade, the position had already been in the red. As Lookonchain stated it’s not surprising, saying Hayes had once again “bought high and sold low.”
Likewise, this contrasts with Hayes’ image of being one of the largest Ethereum long-term optimists. In late 2025, Hayes remarked with confidence:
The coming Ether bull run is about to tear the market a new asshole… Maelstrom is doing all things Ethereum.
Back then, Hayes predicted that ETH would hit the mark of $10,000 by the end of the year.
Since then, however, he has noticeably changed his tone compared to his previous statements. Hayes explained in his article Reality Test:
But right now, it’s about protecting one’s crypto capital.
This statement implies that Friday’s transaction was not so much an exit from Ethereum than a risk-minimizing measure in unpredictable conditions.
The transition is also aligned with Hayes’ overall investing philosophy. Historically, he has made numerous adjustments to his position in response to developments in macroeconomic events, including some high-profile bullish predictions he has made. He has correctly called several major bottom points in the markets, as well as liquidity-driven surges, yet has also accepted when his hypothesis was premature. The importance of the transaction done on Friday resides in its loss, but the readiness to make a change is in line with Hayes’ attitude.
Soft prices and a stretched staking queue
Ethereum still receives conflicting signals. Currently, 41.2 million ETH, which is about 33.8% of the circulating supply, is staked, while the validator activation queue has increased to approximately 43 days, according to data reported by The Block.
Nevertheless, the increasing line of people may not be as optimistic as it looks. Thomas Brunner, the Head of Custody and Staking at Sygnum Bank, informed The Block that most of the activities result from older validators getting their rewards rather than new investors getting into the network.
In addition to its cautious outlook, TD Cowen has recently reduced its end-of-year prediction for the price of Ether to $2,371 from a previous level of around $3,650, as a result of slower-than-anticipated developments concerning the regulations for tokenized assets in the U.S.
In light of that situation, the sale by Hayes has more of a symbolic than a market-impacting value. The trade itself was rather negligible in size that is not enough to influence the price of Ether, but since Hayes is believed to execute trades relying on macro trends rather than on short-term fluctuations, traders understand his moves as insights to the changing corporate risk attitude rather than just another transaction involving a whale.

Don’t just read crypto news. Understand it. Subscribe to our newsletter. It's free.

Solana memecoin platform Pump.fun reportedly terminated over 40 employees shortly before their token allocations vested. The cuts stripped staff of $PUMP tokens, leaving at least one former employee without a seven-figure payout.
The launchpad fired staff in late March and early April. According to a Sandmark report, the firm obtained internal documents, emails, and audio and video recordings of a meeting. Noah Tweedale, one of the company’s co-founders, explained to employees that the cuts were necessary because the firm “had grown too quickly,” preventing Pump.fun from being “fast and rough.”
Pump.fun stayed silent on the allegations
A “termination of services” email, seen by Sandmark, instructed Pump.fun employees to attend the meeting with Lloyd McCarthy, the firm’s head of talent. Employees who lost their jobs had their agreements terminated in early April, and they were compensated with one week of severance pay for each month of service.
Tweedale and his co-founder, Alon Cohen, did not respond to Sandmark’s inquiries. Pump.fun has not issued any response to the allegations.
The employees were terminated due to a two-year delay in the release of their tokens. Pump.fun employees signed token grant agreements in mid-June 2025, and a quarter of their PUMP allocations were set to unlock in June 2026, according to the report. Therefore, those who left the company before that date lost their tokens.
For one employee, this meant giving up on a seven-figure sum. Sandmark noted that, at the time of the report, the value of the allocation was over $1 million at current prices.
The figure holds even though PUMP has fallen sharply. Sandmark, citing TradingView, put the token about 79% below its 2025 peak of $0.0089. Meanwhile, the Crypto Times journal noted a slightly lower drop of 69% from the peak. The token is down 26% per year and changed hands for about $0.002 as the reports circulated, climbing by almost 8% on the day.
Inside Pump.fun’s reported second layoff wave
The ex-employees of Pump.fun noted that there may have been a second round of layoffs, which occurred in mid-July, according to Sandmark. An X account, ExPumpEmployee, began posting “termination of services” emails on Thursday, claiming to have been fired one day before their PUMP tokens would have vested.
Over 40 employees were laid off in two waves, according to both Pump.fun ex-employees, with Sandmark noting that it could not verify the figure or every detail provided. KuCoin’s news desk, to which the report was shared for comment, also indicated that over 40 employees were laid off in two months.
The number is notable given the explosive rise of Pump.fun and its current status as one of crypto’s highest-revenue-generating platforms. The memecoin launchpad saw its user base grow from three to nearly 100 employees this year, as per Sandmark, after attracting users to create over 20.8 million tokens on the platform. It has generated “roughly $1.3 billion in lifetime revenue and still takes in about $1 million per day,” according to the report, which cites analytics from Dune.
A UK-based company with regulatory issues
Pump.fun is incorporated in the UK under the name Baton Corporation Ltd. It has blocked UK users since December 2024, after the Financial Conduct Authority warned it that it may be “carrying on activities which require authorization.” It is currently on the FCA’s warning list.
The UK’s connection to Pump.fun was first established in 2024, when former employee Jarett Dunn transferred around $2 million from one of the company’s wallets to thousands of addresses, claiming to have “killed” the memecoin. Wood Green Crown Court sentenced Dunn to six years in prison in December 2025. Earlier, Cryptopolitan reported that Pump.fun launches bounty feature, and users were posting suicide offers.
Sandmark noted that Baton’s accounts for the period through September 30, 2025, were due by June 30 and were now delinquent, with a potential fine of up to £375. Pump.fun has had several UK-based legal issues, including a lawsuit claiming that it operated a “rigged” offering for investors and another regarding its maximal extractable value practices.

The smartest crypto minds already read our newsletter. Want in? Join them.

General Motors will launch its own in-vehicle artificial intelligence assistant later this year, designed to reach further into the car than the Google Gemini system it started deploying in the spring, per CNBC.
Anna Santos, GM’s director of product management of voice and AI/machine learning, said the assistant “combines conversational AI with GM vehicle knowledge and OnStar intelligence” to do things a general-purpose system cannot. She declined to name the product.
The technology is being developed by GM using a large language model provider whose identity they have not revealed. The features under development include predictive maintenance capabilities, telemetry monitoring and a “kid’s mode” that will lock and unlock doors, change seats, temperature and tunes in one go.
Gemini gives GM a fast AI rollout
GM began the Gemini rollout on April 28 to model year 2022 and newer Cadillac, Chevrolet, Buick, and GMC vehicles with Google built-in, roughly 4 million cars in the United States. Drivers can speak naturally without memorizing commands, and GM has started adding “live sessions” in which the assistant holds looser conversation or plays trivia and 20 questions. It also controls temperature and radio.
Beyond that, CNBC reports, it generally operates the way it would through a phone. That is the gap the native assistant is meant to close, and GM signaled it was coming at its GM Forward event in October 2025 without saying what it would do.
As reported by Cryptopolitan at the time, GM positioned Gemini as the first step towards a dedicated assistant that would work via OnStar.
OnStar data gives GM’s assistant the real edge
What makes the proprietary system proprietary is OnStar telematics, and that is the same asset that drew federal action. The Federal Trade Commission barred GM from sharing OnStar Smart Driver data with insurers and consumer reporting agencies without explicit consent for five years, following reporting that driver behavior data reached insurance companies.
GM says drivers control what Gemini can access. The native assistant sharpens the question rather than settling it, because its entire advantage rests on how much of the vehicle and driver it can see. OnStar reaches vehicles built as far back as 2015, and Google has handled non-emergency OnStar queries through Cloud Dialogflow since 2023.
Automakers are choosing different AI partners
Every major automaker has chosen a partner rather than a build. Stellantis is working with French firm Mistral, Mercedes integrated ChatGPT, and Tesla brought in xAI’s Grok. GM is running both tracks at once, shipping Google’s model to millions of cars while building the system that will sit beneath it.
The business case showed up in the second quarter, when GM reported $48.03 billion in revenue, raised its full-year profit outlook for the second time this year, and posted 20% growth in digital services revenue, per Quartz.
The assistant will also run on hardware GM has been assembling for years, including an Nvidia partnership covering Drive AGX and Omniverse and a centralized computing platform the company says delivers up to 35 times current AI performance, arriving with the Cadillac Escalade IQ in 2028 alongside eyes-off highway driving.

If you're reading this, you’re already ahead. Stay there with our newsletter.

House Select Committee on China Chairman John Moolenaar and Homeland Security Committee Chairman Andrew Garbarino sent DoorDash co-founder and chief executive Tony Xu a letter on Friday requesting information and documents on how the company evaluates and deploys Chinese-developed AI.
House Committees Investigate DoorDash’s Use of Chinese AI: A Recipe for Risk | Source: HoR Press Release
The request grew out of a post on X by DoorDash founder Andy Fang describing the delegation of lower-level AI work to Kimi K2.6, built by Beijing-based Moonshot AI, per CNBC.
DoorDash is not accused of breaking any law. US companies may freely run Chinese models, though federal agencies have blocked some systems including DeepSeek on government devices.
A company spokesperson said DoorDash “proudly supports American AI leadership” and looks forward to engaging with the committees.
Lawmakers say US model pricing is part of the problem
The chairmen conceded the business case in their own text. American firms may deploy Chinese open-weight models because they deliver “competitive capabilities, lower costs, greater customization,” the letter says, and because they offer an alternative to depending on a small number of proprietary providers.
Those considerations, the chairmen wrote, do not remove the need for risk-based safeguards.
The committees claim that “US policies need to ensure that open-weight designs offered by the US become more accessible, safer, and competitive for businesses to discourage them from using Chinese options.”
The committee’s larger argument here is quite straightforward: American businesses and cybersecurity practitioners should not be forced to decide whether they want to use expensive or restrictive US options or cheaper Chinese solutions. The expensive-and-restricted models in this case are Anthropic’s and OpenAI’s.
Open-weight Chinese models raise security questions
The committee’s focus on a specific concern: open-weight models can be downloaded, modified, and deployed by almost anyone because their core model weights are publicly available. The release says these models lack guardrails against bad actors injecting malicious code or launching attacks on other AI systems.
That portability is how Chinese-origin models have reached American enterprise environments, including through cloud services offered by Amazon, Microsoft, and Google. The committees are examining where the models were trained, whether their capabilities came from unauthorized distillation of leading US systems, and how they spread domestically.
As reported previously by Cryptopolitan, the White House science director Michael Kratsios had accused Moonshot of covertly distilling Anthropic’s Fable to develop Kimi K3 based on the report of Anthropic that had revealed more than 3.4 million conversations of Claude from the company using fake accounts.
DoorDash benchmark also points to US model gains
DoorDash’s own numbers explain the pull. Its AI research lab posted on X that both Kimi K2.6 and Anthropic’s Fable 5 outperformed the Anthropic models the company had been running, and cost less doing it. The committees cited the Kimi half. The same benchmark showed an American frontier model beating its predecessor on price and capability, which is the outcome the letter’s own supply argument is asking for.
Washington remains split on open AI rules
On July 24, Nvidia chief executive Jensen Huang used his first post on X to share an open letter signed by Nvidia, Microsoft, Dell, and Palantir arguing that open-weight models strengthen safety, cybersecurity, and sovereignty.
The Little Tech Association, roughly 200 startups including Y Combinator and Proton, and investors Bill Gurley and Ben Horowitz have argued that restricting the category would mostly entrench Anthropic and OpenAI.
The Trump administration has separately weighed barring US companies from running advanced Chinese open-source models, and Treasury Secretary Scott Bessent said on July 21 that the US could sanction China over AI model theft.
The smartest crypto minds already read our newsletter. Want in? Join them.

Former Rep. George Santos agreed Friday to resolve Commodity Futures Trading Commission charges that he manipulated a Kalshi contract on whether he would attend February’s State of the Union address.
This order compels him to disgorge profits in the amount of $17,569.98, a civil monetary penalty of $17,500, cease and desist orders, and trading prohibition for a period of three years, according to the CFTC. He did not admit or deny the findings.
The CFTC found the conduct occurred between February 12 and February 25, and described it as manipulative activity in a contract whose underlying event Santos controlled.
Santos moved the market he was trading
Kalshi users wagered more than $15 million on which political figures would attend the address. Santos took a yes position on himself, then asked followers on X whether he should wear a “muted serious suit to the SOTU or a bedazzled one,” per Forbes. The price of that position climbed, and he exited at a profit, per The Hill.
He ran the sequence repeatedly, narrating a trip toward Washington while trading against the reactions his posts produced. The day before the address, he said in a video that he would be there in the gallery. The following evening, he posted that watching from an airport television had not been the plan.
The CFTC found he acted “willfully or, at the very least, recklessly” and knowingly made misleading public statements to move the contract his way.
Kalshi froze the account and called two regulators
Kalshi detected the trading, froze Santos’s account, and referred it to both the CFTC and the Justice Department, per NOTUS. According to Head of Enforcement Robert DeNault, the firm provided “the evidence needed to bring action against Santos,” and also said that Kalshi intends to take a separate legal course for breaches of exchange rules, as well as try to compensate traders for any losses incurred.
As Cryptopolitan earlier reported, Kalshi made more than 20 referrals to regulators and law enforcement in the first quarter of 2026, opened over 150 insider trading investigations, and blocked more than 100 suspected trades before execution. It has previously banned a California gubernatorial candidate for betting on his own race.
Santos says travel disruption changed his plans
Santos’s attorney, Joseph W. Murray, said his client booked flights and a hotel intending to attend, then reversed after winter storms disrupted East Coast travel and his flight was canceled. Once Santos realized he could not safely attend, Murray said, he adopted a no position, and hid neither the original plan nor the change.
There was absolutely no intent to deceive any person, nor intent to manipulate any market.
Murray.
He noted the contract was the first prediction-market bet Santos had ever placed and that his client cooperated with the agency. Three people with direct knowledge of the trades said the opposite, alleging that Santos deliberately misled the public and profited from it.
CFTC case strengthens Kalshi’s federal oversight argument
The ruling comes amid the commission’s struggle to retain exclusive jurisdiction over the contracts made for events. The CFTC has filed lawsuits against nine states and, on July 14, invoked its emergency powers to stop a Michigan court from compelling Kalshi to void executed trades. Massachusetts, Michigan, Nevada, New York, and Washington have all won rulings restricting the platform.
Friday’s order gives Kalshi a useful example in its wider fight with state regulators. The exchange can now argue that its own monitoring system worked: Kalshi detected the suspicious trading, froze Santos’s account, referred the case to federal regulators, and the CFTC brought an enforcement action.
The case also tests a newer kind of prediction-market manipulation. Santos was not simply trading on public events. He was trading on an outcome he personally controlled, while making public statements that influenced what other traders believed about that outcome.
Santos represented New York in Congress from January 2023 until his expulsion in December 2023. He was later sentenced to 87 months in prison for wire fraud and aggravated identity theft before President Trump commuted the sentence last fall.

If you're reading this, you’re already ahead. Stay there with our newsletter.

An attacker has drained over 1,000 bitcoin (approx worth around $70 million) from nearly 1,200 wallets that used Coldcard hardware to generate a key seed, Galaxy Research revealed on Friday. The figure is substantially higher than the initial estimates made by the company 24 hours earlier. The potential losses involve all Coldcard users who created a seed between March 2021 and the present date.
Galaxy Research estimates that 1,082.65 bitcoin were withdrawn from 1,196 addresses, with the majority of the funds moved between 01:10 and 01:50 UTC on July 30, in a 40-minute window. This number vastly exceeds the reported theft of 594 BTC ($38 million) from approximately 500 single-signature wallets. It happened within 40 minutes. Each of the affected addresses owned more than 0.15 BTC. Over 562 BTC were transferred to one particular address that has yet to see any transactions.
The report highlighted the work of Block’s engineering and security teams, which recognized a pattern in the movement of these funds. “The trail of on-chain transactions identified by other researchers, including Clay Garrett, correctly points to the thief. However, it does not identify the vulnerability that enabled this particular theft,” the company noted, warning that “future attacks could target any Coldcard address generated using the vulnerable firmware.”
How Coldcard’s RNG went wrong?
Coldcard is a hardware wallet that utilizes an air-gapped computer to store the user’s bitcoin. The device employs a cryptographic signing method that should theoretically resist any third-party attempts at interception. It requires a 24-word mnemonic phrase for seed generation that is chosen from a large set of potential words. In reality, the true entropy of a Coldcard wallet was lower than expected.
Block’s Bitcoin Engineering and Security teams published a report that detailed the issue. Within the Coldcard firmware, there were two instances of a function that generated random numbers with the same cryptographic signature, including the hardware implementation written by Coinkite and a software version carried over from MicroPython.
A build-time check that confirmed the presence of the environment setting failed to activate, resulting in some devices utilizing the compromised software random number generator. This vulnerability was patched in the most recent 4.21 release, with the affected versions tracing back to the initial 4.0.0 release from March 2021. The compromised randomness was based on the processor’s serial number and clock, which are not considered secure.
Who is affected by the vulnerability?
The first notice about the security problem came from Coinkite, which published a support page for users who utilized Mk3 hardware with firmware 4.0.1 or newer. The company updated the advisory to also include certain Mk4, Mk5, and Coldcard Q devices and released emergency firmware updates for all affected hardware.
According to reports, Coinkite updated its official advisory to acknowledge that all existing devices are potentially vulnerable to this attack, with firmware upgrades providing only partial security. In theory, all wallets with mnemonic phrases generated before the release of version 4.21 are at risk, according to the Coinkite advisory.
The vulnerability affects more than just seed generation. Coldcard’s paper wallet encryption, key splitting tools, mask generation, and Key Teleport feature all utilize the same randomness function.
Coinkite blames AI for the heist
Coinkite CEO Rodolfo Novak (NVK) published an apology for the security issue, acknowledging that the company was responsible for the error in the mnemonic generation process. “We took full accountability for the firmware bug that led to this situation,” he stated, noting that the initial review of the vulnerability had failed to catch the issue. Novak speculated that the attacker could have used AI to identify the vulnerability, claiming that “this is a sobering reminder of the new paradigm we are entering with AI.”
It is worth noting that Coinkite appears to be arguing against itself, since the Decrypt article notes that the company once used an AI system to scan its own code for potential security problems. That tool was reportedly utilized a few months earlier, with Coinkite claiming that “it did not find this vulnerability or anything else of significance.” Attackers and defenders all have access to the same tools, but in this case, it seems that they failed to do their job.
What should I do if I am a Coldcard wallet owner?
Coinkite recommends that users update the firmware on their devices and create new mnemonic phrases, advising users to make test transactions to the new wallet address before transferring any significant funds. Users should keep their old seed as a backup, as this document will be required to regain access to the funds after the transaction.
The price of bitcoin (BTC) has barely changed in the wake of this revelation. BTC price is hovering near $63,000 on Friday.
If you're reading this, you’re already ahead. Stay there with our newsletter.

Google has removed a new Google Earth feature after roughly a day of operation because disinformation researchers warned that it allowed anyone to drop convincing fake scenes onto Google’s own satellite imagery.
The feature was heavily criticized with several professionals calling for it to be removed. Google attempted to implement safeguards that would allow it to keep the feature operating, but ultimately conceded.
Why did Google approve misinformation?
Google recently launched and quickly removed a Google Earth feature called “create image,” that used Google’s Nano Banana 2 model to turn a typed prompt into an invented scene layered over real satellite, aerial and 3D map data.
A product manager, Bryan Horowitz, said that users could zoom to a spot on the web version, tap the button, and type whatever they wanted to appear there. Google pitched it as a way to “visualize history, create real estate plans and more.”
However, open-source investigator Henk van Ess wrote on Substack, that Google spent 20 years building a trustworthy reference only for it to add a “button that makes things up.”
His main issue with the feature was that the generated images would borrow the authority of the Google Earth platform.
BBC Verify used the tool to generate a collapsed Eiffel Tower, a sinkhole swallowing the Great Pyramid of Giza, and Russian tanks in Kyiv. Van Ess also created images that included refugees at the Mexican border and a bomb crater beside a hospital in Gaza, neither of which met any resistance.
NPR produced images of Iran’s Kharg Island in flames and a flooded U.S. Capitol while AFP fabricated images of an explosion in Paris and a nuclear site in Iran.
404 Media staff added skyscrapers to farmland, planted a “homeless encampment” in Los Angeles, and staged a protest outside Google’s own offices.
Why should satellite imagery be credible?
Open-source analysts rely on images from Google Earth because a photograph taken from hundreds of miles up is costly and difficult to fake.
Jake Godin, a senior researcher at the investigative group Bellingcat, revealed that forged satellite images have surfaced before, including fake damage at a U.S. base in Bahrain during the U.S.-Israel conflict with Iran, but he explained that adding a one-click generator makes the fakes too easy to mass-produce.
Bill Greer, a geospatial analyst and co-founder of the nonprofit Common Space, said the erosion of trust in satellite imagery is “potentially particularly damaging.”
Google explained that every image created with the tool carries a SynthID digital watermark that can be verified through the Gemini app or the Lens feature in Search. It also wrote that it doesn’t allow image creation regarding harmful topics. However, shortly after, BBC Verify found it could trick Gemini into calling a fake Google Earth image genuine.
Google’s prompt filtering also failed easily. While it rejected a request to build a gallows to “hang traitors” outside the UK parliament, a vaguer version of the same prompt went through.
Google has since pulled the feature, stating that while geospatial professionals found the tool useful, people were sharing screenshots of generated imagery that violated its policies. It has stated that it will work on implementing stronger guardrails, but whether the tool will return remains an open question.
The smartest crypto minds already read our newsletter. Want in? Join them.

A federal judge in New York has refused to throw out the core of Reddit’s copyright case against Perplexity AI.
The fight is over how AI search engines harvest user posts and the ruling will set a precedent on whether anti-circumvention law can be used against companies that scrape content to feed AI answers.
What is Reddit accusing Perplexity and SerpAPI of?
U.S. District Judge Paul A. Englemeyer declined to dismiss Reddit’s Digital Millennium Copyright Act claims against Perplexity and the web-scraping platform, SerpAPI. The act makes it illegal to bypass technological measures that control access to copyrighted works, which Reddit claims Perplexity and SerpAPI did using tricks like constantly changing their IP addresses to get around its and Google’s protections in order to grab user posts on a massive scale.
Judge Engelmayer said Reddit’s platform is exactly the kind of “global digital online marketplace for copyrighted works” that the DMCA was designed to protect. He also agreed, for now, that Google’s SearchGuard can count as a “technological measure” that controls access to protected work.
However, the judge dismissed a “trafficking” claim against SerpAPI, along with Reddit’s state-law claims for unfair competition and unjust enrichment against both defendants.
The case will now move into discovery, where Reddit’s claims will be tested against the actual evidence of how Perplexity and SerpAPI collected the posts.
Why are AI companies like Perplexity facing so many lawsuits?
Aside from Reddit, which Cryptopolitan reported filed its lawsuit in October 2025, The New York Times, Dow Jones (which publishes the Wall Street Journal), the New York Post, the Chicago Tribune, CNN, and Encyclopedia Britannica have all filed similar lawsuits against Perplexity. They all accuse the company of using their copyrighted material without permission or payment.
Reddit has said its lawsuit is about protecting its licensing model as it has already made deals to license its content to companies like Google (NASDAQ: GOOGL) and OpenAI for AI training. Reddit argues that Perplexity’s actions undermine these agreements and hurt its business.
Perplexity’s current claim is that it works within the existing rules, while a lawyer for SerpAPI argued that the company “accesses public search results, not Reddit’s platform.” The lawyer added that “public information does not become protected because a platform wants to charge for it.”
Perplexity has compared the lawsuits to how publishers have reacted to every new technology over the last century, from radio to TV to the internet. The company’s head of communications, Jesse Dwyer said, “Fortunately, it’s never worked, or we’d all be talking about this by telegraph.”
Cryptopolitan reported earlier this month that Germany’s ZAK media regulator classified Perplexity AI and Google’s AI Overviews as content providers under German media law, removing a liability shield the companies had relied on.
Perplexity, which is valued at roughly $20 billion, now faces legal pressure on multiple fronts.
Publishers, including the New York Times, the Chicago Tribune, Japan’s Nikkei and Asahi Shimbun, and Encyclopedia Britannica, have all taken the startup to court over copyrighted material.
If you're reading this, you’re already ahead. Stay there with our newsletter.