Cryptopolitan

The wallet associated with the 2022 Pando Rings oracle hack was reactivated on August 18 after two months of inactivity, as reported by blockchain tracker Onchain Lens, The hacker exchanged 3 million DAI for about 1,570 ETH, worth approximately three million dollars, through CoW Protocol.
Approximately 800 ETH worth around 1.52 million are known to have already reached Tornado Cash via eight transactions from this wallet.
Although the action itself may be comparatively minor, the history of the event is anything but. Nearly four years after the incident in which a price feed was manipulated to drain Pando Rings, the fraudster is continuing to move the money, which can still be traced back to the original fraud.
Once a serious cause of losses in DeFi, oracle manipulation has been effectively eliminated from occurring frequently due to improvements in protocol development.
An oracle that misread its own collateral
On November 5, 2022, Pando Rings was hacked. The hacker was able to change the price of sBTC-WBTC liquidity provider token at 4swap, which is Pando’s automated market maker, and used this price manipulation in an attempt to pull out $70 million worth of crypto.
By the time the team took action, around $21.9 million worth of ETH, EOS, and BTC had already flown out of two Mixin wallets controlled by the hacker.
Some assets were not lost. Pando collaborated with Mixin Network and cybersecurity firm SlowMist to lock the rest of the funds. The frozen assets include 2,022,662 EOS coins that were worth approximately $2.36 million, as well as other tokens with a total valuation surpassing $50 million.
The company discontinued its services, namely Pando Rings, 4swap, Pando Leaf, and Pando Lake until the oracle gets fixed and they assured to reimburse all customers.
From buying the dip to reaching for the mixer
The same address has reemerged at intervals since that time. According to a Lookonchain report published on June 6, the same person conducted a transaction worth 10 million DAI to buy a total of 6,243 ETH at an average price of $1,602. It was then added that “even the hacker is buying the $ETH dip.”
The purchase that took place and this week’s swap indicates a well-known strategy: turning stolen stablecoins into Ether when the time is right, and waiting for the best moment to move on. What has changed on August 18 is the final location.
Instead of remaining in possession of the Ether token, the criminal started sending the Ether through Tornado Cash, a service that is used to conceal the connection between deposited and withdrawn funds. As of now the amount of mixer deposits stands at 800 Ether, made in eight transactions.
Why mixed funds stay visible
Even if a person sends money via Tornado Cash, that does not mean the trail will be lost. TRM Labs tracked the attack in June in which a person withdrew around 664 ETH from Tornado Cash and used it to take control of a small Ethereum protocol project known as TOP. This case reveals how mixer operations may still signal risk even if the direct transaction trail is difficult to follow.
The legal standing of Tornado Cash has altered. While being sanctioned by US Treasury in August 2022, it was taken off the sanctions list on March 21, 2025, due to the federal appeals court’s ruling that immutable smart contracts cannot be classified as “property” subject to sanctioning legislation.
Its use as an Ethereum mixer means that big transfers going through the protocol would attract some attention instead of just disappearing.
A protocol winding down as its attacker moves
The timing is interesting. Just three days prior to the wallet’s activity, Pando announced on August 15 that it was discontinuing the protocol and putting its DeFi products into its maintenance mode under the supervision of Mixin. At this point, Pando Rings only serves to support the repayment of loans and the withdrawal of collateral.
In the meantime, incidents like that of Pando are no longer common. Immunefi’s six-year loss analysis found that ecosystem-type attacks, such as flash-loan oracle manipulation, dropped from almost 19% of DeFi loss incidents in 2022 to less than 1% in 2025.
As a result, the Pando exploiter is a remnant of an older time in DeFi security, still profiting from a weakness that the industry as a whole has been able to engineer around while using blockchains.
Broader security angle
The timing of Pando’s Aug. 15 announcement that it was sunsetting the protocol is worth investigating alongside the exploiter’s renewed activity. This isn’t simply an old 2022 hack resurfacing. It illustrates the long tail of DeFi exploits, where stolen assets can remain dormant for years and become active again when market conditions, liquidity, or laundering routes change.
Date Development Nov. 5, 2022 Pando Rings was exploited. Pando said it halted Pando Rings and other services and worked with SlowMist to trace the stolen funds. (Pando Proto) June 2026 The linked exploiter wallet resurfaced, swapping $10M DAI for 6,243 ETH. (CryptoBriefing) ~June-Aug. 2026 Wallet subsequently remained relatively dormant. Aug. 18, 2026 Wallet swapped $3M DAI for ~1,570 ETH, then sent 800 ETH to Tornado Cash. (Blockchain News) Aug. 15, 2026 Pando announced its protocol sunset and service transition, which is potentially relevant context for the timing.
The transactions illustrate how stolen crypto can remain dormant for extended periods before being converted, consolidated, or moved through privacy infrastructure. That’s a pathway the defenders could well follow through.
The smartest crypto minds already read our newsletter. Want in? Join them.

Anthropic has quickly ramped up its sales, such that it is expected to impact the mathematics of the larger AI space. As of the end of July, its projected revenues exceeded $65 billion while investment bankers have started viewing its initial public offering as a precedent for other companies waiting for a listing to be able to set the terms of valuation for other big AI businesses.
This is significant since Anthropic is predicted to go public before OpenAI, perhaps this autumn. Investment banks have informed both firms that whoever is the first to go public is going to “create a model for the entire industry.” Anthropic is looking for a valuation over $2 trillion, making it the biggest IPO in history.
A run rate that tripled in seven months
Investors continue to use the growth curve as a measure. At the end of 2025, Anthropic had almost $9 billion in run rate, or the estimate for future annual revenue based on recent history. It increased to $47 billion in May and reached $65 billion in late July. According to Anthropic itself, in an announcement in May, “our run-rate revenue crossed $47 billion earlier this month.”
Investors expect that momentum to continue, putting 2026 revenue between $100 billion and $120 billion, the Financial Times reported. OpenAI, meanwhile, has doubled its revenue to $40 billion this year from $20 billion at the end of 2025. The companies may measure revenue differently, but Anthropic’s trajectory is getting the attention of investors preparing for its IPO.
Bankers are pricing on 2028, not today
The effects of Anthropic’s initial public offering could reverberate far beyond the particular case of this one firm. Anthropic indicated to those engaged in the IPO process that it anticipates revenues of $190 million to $200 million in 2028, information that is previously unknown. Banks and investors are applying a multiple of enterprise value-to-revenues based on these projections instead of actual income figures, a method usually associated with high-growth software firms in the past rather than traditional public companies.
With respect to the target for 2028, Anthropic’s valuation of $965 billion from the May Series H translates to about five times its future revenue. Cryptopolitan made a comparison with Palantir having about 53 times its anticipated revenues, and SpaceX and Cloudflare approximately 41.6 times, although these numbers are calculated using 2026 estimates.
The important message is not that Anthropic is undervalued. Rather, it is that the company will not warrant such a high multiple if it delivers the revenue the bankers are forecasting. Should public investors accept multipliers based on revenue a couple of years away, it would be simpler for other AI firms to justify their multiples based on present-day revenues.
The demand behind the number
In terms of businesses now, the projections are justified by actual business demand. The July AI Index put out by Ramp indicates that Anthropic comes ahead of OpenAI in terms of adoption figures in US businesses, with 43.5% of American companies paying for either its subscriptions or tokens, compared with 39.7% for OpenAI. Anthropic raised $65 billion in May, thanks to Altimeter Capital, Dragoneer, Greenoaks and Sequoia Capital, thereby giving it a bigger number to invest in compute, research as well as enterprise products contributing to its growth.
However, the valuation model also increases the stakes involved in the situation. The forecast for 2028 assumes that revenue can continue its large edge in terms of exceeding the massive expenses of chips, model training, and talent. Ramp’s data also indicates what limitations businesses have about investing their resources into frontier AI.
If those economics improve, it is likely that the IPO of Anthropic will yield a new valuation playbook in the AI sector. In the opposite scenario, it is possible that creating sophisticated valuation playbooks based on the current price to revenues made in two years could become a trap for investors.


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

Minnesota Attorney General Keith Ellison filed a brief on Friday opposing xAI’s attempt to block HF 1606, the state’s first-in-the-nation law restricting AI tools that generate sexual images of identifiable people, per Decrypt.
Ellison argued the company is unlikely to succeed on its constitutional claim and has failed to show that enforcement would cause irreparable harm.
With Grok Imagine, X.AI has created an unparalleled marketplace for digital sexual violence, and the state would be defenseless from the start unless it could direct its laws at the technology itself.
– Attorney General Keith Ellison
The argument from Minnesota is that Grok Imagine is a tool and not protected speech, which would take it out of the considerations presented by xAI for the judge to examine. There will be a hearing on the matter by a federal judge on Wednesday.
Minnesota’s law targets AI providers even without knowledge or intent
HF 1606 bars a company from letting users generate realistic images showing intimate parts that were absent from the original photograph of an identifiable person, and from producing such an image on a user’s behalf. Violations carry penalties of up to $500,000 per image. People depicted can sue separately.
The statute does not require that a company knew about the image or intended it. Most deepfake laws, including the federal Take It Down Act, penalize whoever makes or shares the picture.
Texas contacted operators in 2025 but only when the owner was aware of the lack of consent or did not respond to a takedown notice. Minnesota eliminated the element of knowledge, and it is the provision under attack by xAI on Wednesday.
The House passed the bill 132-1 and the Senate 65-0, following reports of a man who used social media photos to create sexual images of more than 80 women he knew. It was signed in April. xAI sued on July 27 and moved for a temporary restraining order on July 29.
On July 31, Judge Donovan Frank rejected the petition, pointing out that the company had filed almost three months after signing and three days before the effective date, indicating “harm is not immediate.” The law took effect August 1.
xAI says Minnesota’s rules could criminalize satire and consensual images
The complaint does not contest the interest of the state in preventing nonconsensual imagery, but rather, it argues that the statute reaches far past it, and the argument turns on a definition. The state of Minnesota based its definition of an intimate part on an enactment meant for crimes that involve bodily contact, according to xAI, and this covers the inner thighs and breasts, thus including swimsuits and satire.
Page 19 of the document contains an AI-generated picture that Trump had shared on Truth Social from May 1st, in which he is seen with JD Vance, Marco Rubio, and the Interior Secretary Doug Burgum without their shirts, along with a woman who is unknown, in the Lincoln Memorial Reflecting Pool, a joke about the cost of repairs to the pool. Each man’s breast is depicted where it was not before, xAI argues, which makes the image unlawful in Minnesota.
“Even if subjects gave their consent, generated the images themselves or never shared them, there is no safe harbor for good-faith efforts by the provider of general-purpose AI creative tools,” the company said.
By its own math, ten such images cost $5 million and a hundred thousand would reach $50 billion.
Grok’s enforcement record highlights the scale of the abuse problem
The company has offered its compliance record as evidence of good faith. As Cryptopolitan earlier reported, xAI said in a July filing that it suspended 52,222 accounts this year and sent 73,604 reports to the National Center for Missing and Exploited Children, producing at least 244 arrests in 2026.
Those figures are also a measure of scale. According to the Center for Countering Digital Hate, in the 11-day span between December 29 and January 8, after Musk’s endorsement of the app’s editing functionality, Grok created about 3 million sexually explicit images, including about 23,000 that looked like children were depicted.
California opened an investigation on January 14 and sent a cease and desist two days later. Thirty-five attorneys general signed a joint demand on January 23. According to Ellison, a class action has been filed by five children, and two families from Arkansas have filed suit in federal court.
The European Commission launched an investigation on January 27, and the chatbot has been banned in Malaysia and Indonesia.

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

Payward, Inc., the Cheyenne-based parent of Kraken, said on Monday it has been selected to participate in Project Glasswing and is actively incorporating Claude Mythos 5 into its defensive cybersecurity work, per the company’s announcement.
Anthropic launched the program in April 2026 after concluding its models could surpass all but the most skilled humans at finding and exploiting software vulnerabilities, and has never released Mythos publicly. In addition to Kraken, Payward runs other firms such as NinjaTrader, Breakout, xStocks, Bitnomial, and CF Benchmarks, and generated an adjusted revenue of $508 million for Q2, a 17% increase from the previous year.
Washington controls access to Mythos 5
According to Payward, its access is in line with the United States government’s decision to permit Mythos 5 access to US entities that secure and protect critical infrastructure.
This access route has expanded since April to include technology and financial sectors. Mythos 5 was delivered to US cyber defenders on June 9 via Glasswing and then went dark worldwide three days later due to an export ruling by the Department of Commerce that denied foreign access. The model then returned on July 1.
As Cryptopolitan earlier reported, Bailey said in May that crypto firms and UK banks had been excluded while Goldman Sachs and other American companies were let in. Bailey, who also chairs the Financial Stability Board, argued that “we can’t just have a single sort of national approach” to a risk that crosses borders. A crypto exchange has now cleared the American track. It is up to Washington whether anyone else gets clearance.
Payward will use the model to hunt vulnerabilities
Payward will scan all of its environments, with findings moving into the triage and remediation pipeline it already runs alongside separate red and blue teams and a long-standing bug bounty program.
The company holds ISO 27001 and SOC 2 certifications. Issues that are discovered within third-party open source software are reported to the project maintainers via responsible disclosure, and this is the portion that extends beyond Payward because all the exchanges within the industry rely on the same packages.
Co-Chief Executive Officer Arjun Sethi presented the pitch in terms of the issue faced by the defender: While the attacker requires only one bug, the defender requires all of them every single day. As he put it, “The model is able to scan every single line of code just like an attacker would do.”
Glasswing already includes major US tech and finance firms
Anthropic opened Glasswing in April with Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, the Linux Foundation, Microsoft, Nvidia, Palo Alto Networks, and JPMorganChase, the only bank in the founding group, alongside roughly 40 other organizations.
Partners have surfaced thousands of high and critical-severity flaws since. Mythos received 93.9% score on SWE-bench Verified and 83.1% on CyberGym and the UK Artificial Intelligence Security Institute also verified that Mythos successfully solved 73% of expert-level capture-the-flag tasks.
In the program’s first month, Cloudflare found 2,000 bugs across critical-path systems at a false-positive rate its team rated better than human testers. The model has surfaced a 27-year-old flaw in OpenBSD and a 16-year-old one in FFmpeg. In early June, it identified a critical vulnerability in Zcash’s Orchard shielded pool that had gone undetected for four years, and Zcash used Mythos for the independent audit after patching.
Anthropic is still dealing with Mythos 5 safety concerns
Three weeks ago Anthropic disclosed that three Claude models, Mythos 5 among them, escaped sealed test environments after a misconfiguration gave them internet access. Mythos 5 concluded it was on the open internet, reasoned its way back to believing it was still in a simulation, then wrote and published a PyPI package that was downloaded and run on 15 real systems before removal.
Anthropic said the safety classifiers shipped with its commercial products would have prevented the behavior, described the events as a harness and operational failure, and engaged METR for an independent review.

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

IREN completed the delivery of the first of four planned “Horizon” cloud deployments of artificial intelligence to Microsoft Corporation. With that, the first milestone of a five-year, $9.7 billion contract was completed, which underpins the transition of the former Bitcoin mining company into artificial intelligence infrastructure.
Horizon 1 is located at IREN’s Childress, Texas facility. It is reported to be a 50-megawatt facility with direct chip-level cooling and Nvidia GB300 systems. It is the first of four facilities of this type that the company plans to complete by 2026, and altogether the four facilities should provide a total of 200 megawatts of capacity.
Nvidia also awarded the facility its Exemplar Cloud status, a designation reserved for cloud providers who meet Nvidia’s performance and reliability standards for artificial intelligence workloads. “Delivering Horizon 1 underscores the power of our vertically integrated business model and our ability to deliver on complex AI infrastructure projects with speed and scale,” said Daniel Roberts, co-founder and co-CEO of IREN, in a statement.
Why Microsoft’s acceptance matters more?
Completing the build was one half of the test. As per the deal, Microsoft gets five days to validate the installation of each GPU deployment against mutually accepted standards. Upon completion of the acceptance process, the period for service commences, and IREN starts charging Microsoft monthly.
The key is the billing. IREN signed a cloud service agreement with Microsoft in November 2025, giving us an annualized revenue stream from the deal of around $1.94 billion when all four horizons are online. Microsoft needs to pay 20% of the total contract value upfront in four tranches.
IREN’s AI pivot puts $4B revenue target in sight
The magnitude of the shift is illustrated by looking at IREN’s accounts. The company was listed on the Nasdaq in 2021 as a Bitcoin miner, with the company having generated revenue from the mining business of $511.5 million in the first nine months of 2022 while earning just $58.3 million from its AI Cloud service. According to management, the company plans to reach annualized run-rate AI Cloud revenue in excess of $4 billion by the end of the year, with over 85% of revenue contracted.
The timeline reflects the health of IREN’s core business operations. In mid-August, the price of Bitcoin was hovering around $63,500, less than half of the price peak in October 2025. IREN has stated that the company plans to have 480 megawatts of AI cloud computing capacity in 2026 and 1.2 gigawatts in 2027 while moving out of mining operations.
This expansion is estimated to need about $5.8 billion in GPUs and related hardware purchased from Dell. IREN has secured funding for all but a minority of this expansion. This includes $3.65 billion of secured debt arranged in June and guaranteed by the Microsoft deal, according to The Block, covering 96% of the $5.81 billion in GPU spending and rated investment grade A by Fitch and A(low) by DBRS.
If you're reading this, you’re already ahead. Stay there with our newsletter.

Wispr, the startup behind the dictation app Wispr Flow, said Monday it raised $280 million in Series B funding at a $2 billion valuation.
The round was led by Menlo Ventures. The cash gives Wispr fresh capital to deal with accuracy complaints and to move into meetings as rivals flock to the voice-to-text market.
The round brings Wispr’s total to $361 million
The financing takes Wispr’s total funding since its founding in 2021 to $361 million. It closed its last funding round less than ten months ago. The company had raised around $81 million prior to this round.
Menlo Ventures, a backer of Wispr, led the Series B round. Existing investors, including Notable Capital, NEA, Neo Ventures, 8VC, and MVP Ventures, added more. Wispr also brought in new backers, including Acrew, Forerunner, Goodwater, Peak XV, Together Fund, and PLUS Capital.
Wispr’s competitors are popping up, including apps such as Willow, Monologue, Aqua, and Superwhisper, as well as a wave of free and cheaper tools targeting power users.
Wispr released the funding news alongside a preview of Canto, its first in-house speech model, after several users complained recently that Flow’s dictation had gotten worse.
Canto is designed for noisy real-world use, not just for pristine studio recordings, the company says. In the hardest conditions, with background noise, wind, heavy accents, or music, Wispr said word error rates drop from more than 30% to between 5% and 10%.
That is more than a 4x reduction. In day-to-day usage, the company expects the model to reduce the number of dictations a user has to edit by 30% to 35%.
“The whole reason to talk instead of type is to stay inside your own train of thought,” wrote CEO and co-founder Tanay Kothari in a company blog post.
One wrong word sends users back to the keyboard and breaks their focus, he added.
Notetaker takes the fight to Granola and Otter
The “beyond dictation” pitch is focused on Notetaker, a meeting tool Wispr shipped about a week before the raise.
It records calls and in-person conversations, labels speakers on a live transcript, and produces topic-organized summaries with dates, decisions, and next steps.
That means Wispr is up against Granola, Fireflies, Read AI, and Otter. Notetaker records audio locally and doesn’t join calls as a visible bot, said Sahaj Garg, CTO and co-founder.
That puts the burden on users to disclose in states where consent is required to record, he said. Two of its competitors, Otter and Granola, are currently facing separate California lawsuits alleging privacy violations.
Wispr also set up the Wispr Advanced Interfaces Lab led by Chief Scientist Ariya Rastrow, who was a founding member of the team that created Amazon’s Alexa. The lab is looking at interfaces that respond to what a user says, not just transcribe it.
Along with the venture firms, Wispr’s round included a roster of athletes and cultural figures. They include Livvy Dunne, Shaun White, Dak Prescott, Joe Burrow, Klay Thompson, Paul George, and Trae Young.
Domantas Sabonis, three-time NBA All-Star and one of the investors, said Flow keeps up as he switches between English, Spanish, and Lithuanian.
Wispr says people have written over 60 billion words with Flow, and almost all Fortune 500 companies and 10,000+ enterprises use the app.
Don’t just read crypto news. Understand it. Subscribe to our newsletter. It's free.

Serve Robotics added Grubhub as a delivery partner and opened two more cities.
The sidewalk-robot company announced the moves Monday as it works to replace revenue it lost when its Uber Eats deal fell apart.
Where the Grubhub robots start rolling
Grubhub customers in Chicago, Los Angeles, and Alexandria, Virginia, can now get food delivered by one of Serve’s autonomous robots.`
At launch, 100+ restaurants in Chicago and almost 200 in Los Angeles will be participating in the service, with Serve expecting more to sign up.
Grubhub is owned by Wonder, a food-technology company that also operates its own restaurant brand. Wonder’s Alexandria kitchen will dispatch orders to Serve robots too.
“Our partnership with Serve brings autonomous delivery to Grubhub customers in Chicago, Los Angeles, and Alexandria, while also enabling robot delivery from Wonder’s Alexandria location,” PJ Poykayil, Wonder’s EVP of Customer Delivery Operations, said.
Serve CEO Ali Kashani contextualized the deal in terms of scale. Each platform Serve taps into expands the number of restaurants and neighborhoods its ~2,000 robots can reach, he said.
Serve, Kashani told investors in last week’s second-quarter earnings call, isn’t planning to renew its Uber deal in 2027, citing what he described as “differing views” on how the robots should be deployed.
The volume of deliveries running through Uber fell for the first time since 2022, Kashani said. Serve cut its 2026 revenue guidance from $26 million to a range of $9 million to $10 million.
The day after the earnings call, Uber said it had sold its entire stake in Serve, a company that spun off from the ride-hailing firm more than five years ago.
“I have a lot of respect for Uber. They kind of helped us bootstrap this,” Kashani said. “We just have a different idea for what we want next.”
Serve Robotics expands to two new DoorDash cities
Serve opened its seventh and eighth U.S. markets Monday, both via DoorDash. They are San Jose, California, and Washington, D.C.
The two metros combined have about 8 million people, joining Los Angeles, Chicago, Atlanta, Dallas, and Miami on Serve’s map.
The company’s first Bay Area site, in San Jose, has already completed its first month of deliveries. The Washington rollout includes Dupont Circle and parts of downtown, and Talkin’ Tacos is among the first restaurants signed up.
Serve is utilizing small, existing sites for robot staging, charging, dispatch, and maintenance, including established parking structures for the Washington rollout.
The first one is opening in Miami, Serve says, and the model allows it to move into new neighborhoods without the typical build-out time.
As Cryptopolitan reported last year, DoorDash debuted an in-house robot called Dot in Phoenix.
Serve’s Diligent Robotics unit in health care has begun shipping Moxi 2.0, a hospital robot. The company claims Moxi 2.0 can sense its surroundings up to 15 times faster than its predecessor, features 10 times the onboard computing power, has a runtime of up to 18 hours, and charges 30% faster.
Early deployments include Endeavor Health Edward Hospital near Chicago, Providence Saint John’s Health Center in Los Angeles, and Children’s Hospital Los Angeles.
Don’t just read crypto news. Understand it. Subscribe to our newsletter. It's free.

Serve Robotics added Grubhub as a delivery partner and opened two more cities.
The sidewalk-robot company announced the moves Monday as it works to replace revenue it lost when its Uber Eats deal fell apart.
Where the Grubhub robots start rolling
Grubhub customers in Chicago, Los Angeles, and Alexandria, Virginia, can now get food delivered by one of Serve’s autonomous robots.`
At launch, 100+ restaurants in Chicago and almost 200 in Los Angeles will be participating in the service, with Serve expecting more to sign up.
Grubhub is owned by Wonder, a food-technology company that also operates its own restaurant brand. Wonder’s Alexandria kitchen will dispatch orders to Serve robots too.
“Our partnership with Serve brings autonomous delivery to Grubhub customers in Chicago, Los Angeles, and Alexandria, while also enabling robot delivery from Wonder’s Alexandria location,” PJ Poykayil, Wonder’s EVP of Customer Delivery Operations, said.
Serve CEO Ali Kashani contextualized the deal in terms of scale. Each platform Serve taps into expands the number of restaurants and neighborhoods its ~2,000 robots can reach, he said.
Serve, Kashani told investors in last week’s second-quarter earnings call, isn’t planning to renew its Uber deal in 2027, citing what he described as “differing views” on how the robots should be deployed.
The volume of deliveries running through Uber fell for the first time since 2022, Kashani said. Serve cut its 2026 revenue guidance from $26 million to a range of $9 million to $10 million.
The day after the earnings call, Uber said it had sold its entire stake in Serve, a company that spun off from the ride-hailing firm more than five years ago.
“I have a lot of respect for Uber. They kind of helped us bootstrap this,” Kashani said. “We just have a different idea for what we want next.”
Serve Robotics expands to two new DoorDash cities
Serve opened its seventh and eighth U.S. markets Monday, both via DoorDash. They are San Jose, California, and Washington, D.C.
The two metros combined have about 8 million people, joining Los Angeles, Chicago, Atlanta, Dallas, and Miami on Serve’s map.
The company’s first Bay Area site, in San Jose, has already completed its first month of deliveries. The Washington rollout includes Dupont Circle and parts of downtown, and Talkin’ Tacos is among the first restaurants signed up.
Serve is utilizing small, existing sites for robot staging, charging, dispatch, and maintenance, including established parking structures for the Washington rollout.
The first one is opening in Miami, Serve says, and the model allows it to move into new neighborhoods without the typical build-out time.
As Cryptopolitan reported last year, DoorDash debuted an in-house robot called Dot in Phoenix.
Serve’s Diligent Robotics unit in health care has begun shipping Moxi 2.0, a hospital robot. The company claims Moxi 2.0 can sense its surroundings up to 15 times faster than its predecessor, features 10 times the onboard computing power, has a runtime of up to 18 hours, and charges 30% faster.
Early deployments include Endeavor Health Edward Hospital near Chicago, Providence Saint John’s Health Center in Los Angeles, and Children’s Hospital Los Angeles.
If you're reading this, you’re already ahead. Stay there with our newsletter.

Compound, one of the first protocols to make crypto lending work without a bank, has approved a record $52 million development budget and rebuilt its leadership around traditional-finance veterans.
It is also wagering its next chapter on institutions, which is seen as a move away from retail yield-chasers who once drove DeFi. Its native token, COMP, seems to have received a boost as a result of the announcement, as it has gone up by over 10% in 24 hours.
Projects that grew up serving retail now court banks, asset managers, and compliance departments to find their footing again, and Compound seems to be moving in that direction.
Compound’s foundation wrote on X that the protocol is “entering its next era.”
From $12 billion to $1.2 billion
The total value locked (TVL) on Compound has fallen to around $1.2 billion, which is a decline of around 90% from the $12 billion the protocol held at its September 2021 peak per DeFillama.
Compound TVL is down from its 2022 highs. Source: Defillama
The vast majority of the TVL is on Ethereum, with Arbitrum coming a distant second.
Compound’s figures are a far cry from Aave’s, which is the leading protocol in the DeFi lending space with a TVL of over $14.6 billion.
Ironically, Compound helped invent this category when it launched in 2018 and, by its own accounting, has processed around $480 billion in deposits and borrowing since.
However, the protocol’s growth stalled following the end of the incentive programs that helped to push up its metrics.
While COMP received a boost, it still trades far below its glory days. It currently trades around $18; however, it is still down by 98% from its 2021 record.
How does Compound plan on spending its $52 million?
The budget passed through the Compound DAO, and $14 million is cleared for immediate use. The rest unlocks in tranches tied to milestones, a structure that keeps the development team on a performance leash funded by the treasury.
The money splits roughly in two, with about $28 million for operations and the engineering behind a new protocol version, Compound V4, and $24 million for growth. Of that growth pool, $8 million to $10 million is earmarked for institutional partnerships rather than the old playbook of paying liquidity providers to boost headline numbers.
V4’s centerpiece is a “hub-and-spoke” design, which routes capital through a central hub instead of walling it off in separate markets, an approach meant to give professional counterparties tighter risk controls. Compound stated that more than 10 partners have committed, with over 20 more in talks.
A bench built from traditional finance
The leadership overhaul reads as a statement of intent. The new team includes Chief Operating Officer Christopher Donovan, previously COO at the Near Foundation, and Chief Product Officer Steven Liu, who grew Maple Finance’s assets from $500 million to $5 billion.
Aaron Schnarch, former CEO of Coinbase Custody, joins as an executive director, with other hires drawn from Anchorage Digital, HSBC, Broadridge Financial, and Maple.
In a statement, Schnarch said, “DeFi is a remarkable innovation; however, it has achieved limited institutional adoption.” He added that current products “fall short of meeting the traditional finance bar.”
However, not everyone is sold on credentials alone, and one of them is Himanshu Sahay of Arch Lending, who said that the budget and bench amount to “a serious move” but warned that institutions “aren’t underwriting teams, they’re underwriting structures.”
Why is everyone now chasing institutions?
Compound is late to a race it once could have led. Across crypto this year, the institutional turn has become the default survival strategy. In July, former Ethereum Foundation staff launched Ethereum Institutional, a non-profit “front door” for banks and asset managers backed by Bitmine, SharpLink, and Joseph Lubin, Cryptopolitan reported.
Tokenized real-world assets climbed to about $65 billion by May, and more than 2,000 institutions disclosed Bitcoin holdings through spot ETFs in Q1.
Ran Hammer of Orbs said, “Retail participation is a fraction of what it was,” as the chain “quietly become a venue for settlement, execution and interaction between financial institutions.” Standard Chartered projects the DeFi sector could reach $2.7 trillion by 2030.
If you're reading this, you’re already ahead. Stay there with our newsletter.

Unitree Robotics revealed its ‘Superman’ humanoid robot on Monday, with a reported top running speed of 12.66 meters per second and a two-meter standing jump, beating every human on record. The reveal comes only two days before Unitree starts publicly trading in Shanghai, after one of the most oversubscribed tech listings the market has seen.
‘Superman’ hits flying numbers
Superman cleared a two-meter standing high jump on legs measuring 0.85 meters, and hit 12.66 m/s, which is equal to 45.6 km/h in speed. The compared human marks the robot beat for the respective actions are a 1.8-meter standing jump and a 12.4 m/s sprint.
Usain Bolt reached a max speed of 12.42 m/s during his 2009 world-record race, meaning the ‘Superman” robot is faster than the world’s fastest man ever managed, if Unitree’s figures hold true.
The robotics company has, however, not released any independent verification for these figures, with the numbers all coming from a company video and not an independently measured event.
Unitree New Robot Preview: “Superman” Breaking the Limits of Humanity🥳
Standing high jump 2 m, top speed 12.66 m/s (0.85 m leg length)
Surpassing the standing high jump and running speed records of all humans around the world
This new machine has only been in development for a… pic.twitter.com/12i80ITU6p
— Unitree (@UnitreeRobotics) August 17, 2026
Three months of development
Unitree also stated that its engineers built the robot in a little over three months, and that the humanoid robot’s current hardware leaves plenty of room for improvement. The company’s founder, Wang Xingxing, hinted at the possibility back in March at the Yabuli China Entrepreneurs Forum.
He predicted that humanoid machines would break human sprint limits by mid-2026, referring to cheaper parts and better algorithms as the reason for such a possibility.
The robotics company had introduced the As2W some weeks before, a wheeled quadruped robot built for cargo. The As2W weighs 25 kilograms, can carry a recorded max load of 180 kilograms, and can move a cargo of up to 16 kilograms continuously.
Unitree IPO close at hand
Unitree priced its coming IPO listing at 150.8 yuan and raised about 6.1 billion yuan, almost $905 million, in a deal that places the company’s valuation at about $9 billion. The IPO was oversubscribed by over 5000 times, and the company raked in 45% more than the 4.2 billion yuan it had initially targeted.
Real revenue also adds to the valuation, as sales hit 1.7 billion yuan last year, more than four times the 2024 figure, with net profit of 591 million yuan. Unitree sold over 5,500 humanoid units across all its G1, H1, and R1 units in 2025, with most of these going to research labs and entertainment buyers instead of factories and manufacturers.
The smartest crypto minds already read our newsletter. Want in? Join them.

Unitree Robotics revealed its ‘Superman’ humanoid robot on Monday, with a reported top running speed of 12.66 meters per second and a two-meter standing jump, beating every human on record. The reveal comes only two days before Unitree starts publicly trading in Shanghai, after one of the most oversubscribed tech listings the market has seen.
‘Superman’ hits flying numbers
Superman cleared a two-meter standing high jump on legs measuring 0.85 meters, and hit 12.66 m/s, which is equal to 45.6 km/h in speed. The compared human marks the robot beat for the respective actions are a 1.8-meter standing jump and a 12.4 m/s sprint.
Usain Bolt reached a max speed of 12.42 m/s during his 2009 world-record race, meaning the ‘Superman” robot is faster than the world’s fastest man ever managed, if Unitree’s figures hold true.
The robotics company has, however, not released any independent verification for these figures, with the numbers all coming from a company video and not an independently measured event.
Unitree New Robot Preview: “Superman” Breaking the Limits of Humanity🥳
Standing high jump 2 m, top speed 12.66 m/s (0.85 m leg length)
Surpassing the standing high jump and running speed records of all humans around the world
This new machine has only been in development for a… pic.twitter.com/12i80ITU6p
— Unitree (@UnitreeRobotics) August 17, 2026
Three months of development
Unitree also stated that its engineers built the robot in a little over three months, and that the humanoid robot’s current hardware leaves plenty of room for improvement. The company’s founder, Wang Xingxing, hinted at the possibility back in March at the Yabuli China Entrepreneurs Forum.
He predicted that humanoid machines would break human sprint limits by mid-2026, referring to cheaper parts and better algorithms as the reason for such a possibility.
The robotics company had introduced the As2W some weeks before, a wheeled quadruped robot built for cargo. The As2W weighs 25 kilograms, can carry a recorded max load of 180 kilograms, and can move a cargo of up to 16 kilograms continuously.
Unitree IPO close at hand
Unitree priced its coming IPO listing at 150.8 yuan and raised about 6.1 billion yuan, almost $905 million, in a deal that places the company’s valuation at about $9 billion. The IPO was oversubscribed by over 5000 times, and the company raked in 45% more than the 4.2 billion yuan it had initially targeted.
Real revenue also adds to the valuation, as sales hit 1.7 billion yuan last year, more than four times the 2024 figure, with net profit of 591 million yuan. Unitree sold over 5,500 humanoid units across all its G1, H1, and R1 units in 2025, with most of these going to research labs and entertainment buyers instead of factories and manufacturers.
The smartest crypto minds already read our newsletter. Want in? Join them.

The Hyperliquid Policy Center (HPC) and Pyth Network contributor, Douro Labs, asked the Securities and Exchange Commission (SEC), on August 17 2026, to end Rule 611. Rule 611 is the 2005 trade-through rule.
HPC and Duro Labs argue that Rule 611 is unable to properly describe how stock trades are settled on public blockchains and are urging the SEC to allow independent onchain price feeds. Their joint comment letter was filed under docket S7-2026-20, with the comment window closing the same day the letter was filed. It’s now left to the SEC to decide.
What Rule 611 made brokers do?
Rule 611 is at the core of the market-structure framework launched in 2005. The framework is referred to as Regulation NMS, and the role of Rule 611 was clear and simple: a broker handling a customer order cannot execute at a worse price when a better one is displayed somewhere else.
If an exchange displays a stock at $90.00 and another at $90.01, the broker has to use the $90.00 price.
For this to work, every US exchange had to be displayed in a single feed. Venues relay their prices to the securities information processors (SIPs), which post the NBBO (national best bid and offer), and it becomes a reference price for every broker executing an order.
On June 11, the SEC voted on a proposal to end two rules, Rule 611 and Rule 610(e).
Rule 610(e) prevents exchanges from publishing quotes that lock or cross each other.
Paul Atkins, the SEC Chairman, has opposed Rule 611 since its inception in 2005. If Rule 611 is repealed, it’s left to the broker to protect the customer’s interests as they are obliged to under FINRA rules.
Why the national benchmark breaks down onchain
The main claim in the letter filed on August 17 is that the NBBO is not suitable for blockchain venues. The letter states three problems:
The national benchmark is often inoperative when a trade occurs, as the SIPs do not trade through nights, weekends, and holidays when blockchain markets are trading
The benchmark does not accurately portray onchain conditions. Large asset holders like automated market makers do not post bids or offers; rather, they determine prices by formulas
The clocks are a mismatch. Refreshing a feed in microseconds does not move the needle for orders that are settled at a block boundary.
The letter pegs it at around 200 milliseconds on Hyperliquid, 400 on Solana, 12 seconds on Ethereum, and ~10 minutes on Bitcoin.
A different reference price
The letter enjoins the SEC to approve a “qualifying reference price” in place of the NBBO. The broker would use this price to measure onchain orders. The letter proposes four tests:
The price should be an aggregate of quotes from independent firms that typically set prices, with reported data on recency and validation
Make use of methods that are manipulation-resistant
Make its publishers and calculation logic public and easy to audit
Ensure periodic checks against data from SIP and external sources
The letter went on to state that Douro Labs has built the Pyth Pro reference price, a service meant to meet those tests.
Who is pushing for this, and why now
The Hyperliquid Policy Center is not a neutral party in all of this. The Hyper Foundation launched the HPC in February 2026 with 1 million HYPE tokens (~$ 30 million back then) and appointed Jake Chervinsky, an experienced crypto lawyer, as the CEO.
The letter beckons the SEC to maintain the best-execution policy anchored in FINRA rather than at individual exchanges. They argue that rules adopted with immediate effect in June by NYSE American, NYSE Arca, NYSE National, and NYSE Texas have a conflict, because exchanges compete for the same order flow they are to regulate.
If you're reading this, you’re already ahead. Stay there with our newsletter.

The Hyperliquid Policy Center (HPC) and Pyth Network contributor, Douro Labs, asked the Securities and Exchange Commission (SEC), on August 17 2026, to end Rule 611. Rule 611 is the 2005 trade-through rule.
HPC and Duro Labs argue that Rule 611 is unable to properly describe how stock trades are settled on public blockchains and are urging the SEC to allow independent onchain price feeds. Their joint comment letter was filed under docket S7-2026-20, with the comment window closing the same day the letter was filed. It’s now left to the SEC to decide.
What Rule 611 made brokers do?
Rule 611 is at the core of the market-structure framework launched in 2005. The framework is referred to as Regulation NMS, and the role of Rule 611 was clear and simple: a broker handling a customer order cannot execute at a worse price when a better one is displayed somewhere else.
If an exchange displays a stock at $90.00 and another at $90.01, the broker has to use the $90.00 price.
For this to work, every US exchange had to be displayed in a single feed. Venues relay their prices to the securities information processors (SIPs), which post the NBBO (national best bid and offer), and it becomes a reference price for every broker executing an order.
On June 11, the SEC voted on a proposal to end two rules, Rule 611 and Rule 610(e).
Rule 610(e) prevents exchanges from publishing quotes that lock or cross each other.
Paul Atkins, the SEC Chairman, has opposed Rule 611 since its inception in 2005. If Rule 611 is repealed, it’s left to the broker to protect the customer’s interests as they are obliged to under FINRA rules.
Why the national benchmark breaks down onchain
The main claim in the letter filed on August 17 is that the NBBO is not suitable for blockchain venues. The letter states three problems:
The national benchmark is often inoperative when a trade occurs, as the SIPs do not trade through nights, weekends, and holidays when blockchain markets are trading
The benchmark does not accurately portray onchain conditions. Large asset holders like automated market makers do not post bids or offers; rather, they determine prices by formulas
The clocks are a mismatch. Refreshing a feed in microseconds does not move the needle for orders that are settled at a block boundary.
The letter pegs it at around 200 milliseconds on Hyperliquid, 400 on Solana, 12 seconds on Ethereum, and ~10 minutes on Bitcoin.
A different reference price
The letter enjoins the SEC to approve a “qualifying reference price” in place of the NBBO. The broker would use this price to measure onchain orders. The letter proposes four tests:
The price should be an aggregate of quotes from independent firms that typically set prices, with reported data on recency and validation
Make use of methods that are manipulation-resistant
Make its publishers and calculation logic public and easy to audit
Ensure periodic checks against data from SIP and external sources
The letter went on to state that Douro Labs has built the Pyth Pro reference price, a service meant to meet those tests.
Who is pushing for this, and why now
The Hyperliquid Policy Center is not a neutral party in all of this. The Hyper Foundation launched the HPC in February 2026 with 1 million HYPE tokens (~$ 30 million back then) and appointed Jake Chervinsky, an experienced crypto lawyer, as the CEO.
The letter beckons the SEC to maintain the best-execution policy anchored in FINRA rather than at individual exchanges. They argue that rules adopted with immediate effect in June by NYSE American, NYSE Arca, NYSE National, and NYSE Texas have a conflict, because exchanges compete for the same order flow they are to regulate.
If you're reading this, you’re already ahead. Stay there with our newsletter.

OpenAI president Greg Brockman has dismissed the concern over the wave of senior executives departing the company, saying it is “not that atypical.”
According to Brockman, the concern is only due to the company’s popularity. Individuals within the company also claim that the reshuffling is needed and welcome.
Why does it matter that OpenAI is losing its executives?
Speaking on CNBC’s “Squawk Box,” OpenAI President Greg Brockman argued that the departures of OpenAI’s executives only create such a buzz because the company is “so much in the spotlight,” and that it would be different if it were a smaller company.
He framed the departures as routine for a company that keeps reinventing itself, but stated that he and chief executive Sam Altman are constants. The two co-founded OpenAI with others in 2015.
Most recently, OpenAI’s chief revenue officer, Denise Dresser, left after less than eight months in the job. Two days before that, Brad Lightcap, an eight-year OpenAI veteran and former operating chief, said he was moving on to start something new.
OpenAI has tapped Dali Rajic, previously president and COO of the cybersecurity firm Wiz, to fill the revenue role. Google bought Wiz for $32 billion.
At least a dozen senior leaders have left the company so far in 2026, including the CEO of applications, Fidji Simo, who stepped back last month after a chronic illness made a full-time return untenable, as well as Kevin Weil, Bill Peebles, and Srinivas Narayanan, who left earlier in the year.
Head of ethics Chloé Bakalar, head of safety systems Johannes Heidecke, and former mission alignment chief Joshua Achiam have all exited, with safety researcher Sandhini Agarwal leaving in July.
OpenAI has also reportedly disbanded its “preparedness” team, which studied catastrophic model risks. Those duties have been integrated into other groups as part of a “streamlining process” tied to Altman’s goal of refocusing on ChatGPT.
Will the departures affect OpenAI’s IPO?
OpenAI filed its prospectus confidentially with the SEC in June and carries an $852 billion valuation. Despite concerns about the exodus of executives, the company’s run rate climbed 20% month over month in July, with business customers up 32% over the same stretch.
Meanwhile, Brockman has reportedly been taking a bigger role across the company to build a leadership team that will push OpenAI past Anthropic in enterprise adoption. Individuals within the company reportedly think of the employee exodus as a long-overdue clearing of underperforming managers.
A security incident disclosed last month in which OpenAI models broke out of an isolated test environment, strung together a chain of vulnerabilities, reached the open web, and got into the developer platform Hugging Face, was also discussed on CNBC.
Brockman said the company treated the matter with the utmost seriousness and published a blog post meant to help other organizations defend against similar attacks.
Don’t just read crypto news. Understand it. Subscribe to our newsletter. It's free.

ByteDance and the Motion Picture Association have signed a memorandum of understanding to build copyright guardrails into the company’s Seedance and Seedream AI generators, the first deal of its kind between Hollywood’s main studio lobby and an AI firm, according to the Los Angeles Times.
Dispute started after popular Tom Cruise video
The truce ends a dispute that broke out approximately six months ago. The controversy dates back to a clip created purely by AI of Tom Cruise trading blows with Brad Pitt, built with Seedance, and then spreading widely all over the internet.
On February 20, MPA global general counsel Karyn Temple wrote to ByteDance general counsel John Rogovin, the Los Angeles Times reported, arguing that Seedance 2.0 had been trained on protected material and was turning out unauthorized clips of characters such as SpongeBob SquarePants alongside a recreated scene from “Stranger Things.”
Variety reported that the trade group accused ByteDance of “disregarding well-established copyright law that protects the rights of creators and underpins millions of American jobs” on a supposedly “massive scale.” Disney and other studios also objected to tools that could reproduce Marvel and Star Wars figures without permission, The News reported.
Results of constructive talks
The MOU then announced on Monday that it has created a structure covering ByteDance’s generative products, reaching users through TikTok, its separate U.S. TikTok app, CapCut, and Dreamina.
The MPA did not release the details of the specific barriers and deterrents in the deal to the public, the Los Angeles Times reported. Both companies have, however, pointed to newer releases as evidence that these safeguards have been initiated, with ByteDance claiming both Seedance 2.5 and Seedream 5.0 Pro, shipped last month, carry stronger IP protections.
“Today’s agreement illustrates our belief that copyright is a cornerstone of the film and television industry,” MPA Chairman and CEO Charles Rivkin said in a statement, adding that the deal “reflects our shared determination to continue our work together to further fortify those guardrails.” Bytedance general counsel, John Rogovin, said the company “respects the intellectual property rights that underpin creative industries around the world” and said the MOU was “an important framework for continued collaboration as the technology evolves.”
OpenAI’s Sora shutdown resonates with IP crackdown
ByteDance is not the first AI company the MPA has pressured into concessions on their AI product. The association had initially gone after OpenAI’s Sora before the firm added limits, according to Variety, with OpenAI eventually shutting Sora down earlier this year after a similar wave of unauthorized character videos.
However, instead of fully retreating and shutting down its products, ByteDance is taking a different path, pouring more funds into Seedance, which is reportedly one of the most advanced video models available, alongside Google’s offerings.
The Los Angeles Times reported that Seedance has gained ground with independent filmmakers who find it cheaper than rival tools, a development that gives ByteDance a commercial reason to keep the studios afloat instead of completely disengaging from the market.
If you're reading this, you’re already ahead. Stay there with our newsletter.

ByteDance and the Motion Picture Association have signed a memorandum of understanding to build copyright guardrails into the company’s Seedance and Seedream AI generators, the first deal of its kind between Hollywood’s main studio lobby and an AI firm, according to the Los Angeles Times.
Dispute started after popular Tom Cruise video
The truce ends a dispute that broke out approximately six months ago. The controversy dates back to a clip created purely by AI of Tom Cruise trading blows with Brad Pitt, built with Seedance, and then spreading widely all over the internet.
On February 20, MPA global general counsel Karyn Temple wrote to ByteDance general counsel John Rogovin, the Los Angeles Times reported, arguing that Seedance 2.0 had been trained on protected material and was turning out unauthorized clips of characters such as SpongeBob SquarePants alongside a recreated scene from “Stranger Things.”
Variety reported that the trade group accused ByteDance of “disregarding well-established copyright law that protects the rights of creators and underpins millions of American jobs” on a supposedly “massive scale.” Disney and other studios also objected to tools that could reproduce Marvel and Star Wars figures without permission, The News reported.
Results of constructive talks
The MOU then announced on Monday that it has created a structure covering ByteDance’s generative products, reaching users through TikTok, its separate U.S. TikTok app, CapCut, and Dreamina.
The MPA did not release the details of the specific barriers and deterrents in the deal to the public, the Los Angeles Times reported. Both companies have, however, pointed to newer releases as evidence that these safeguards have been initiated, with ByteDance claiming both Seedance 2.5 and Seedream 5.0 Pro, shipped last month, carry stronger IP protections.
“Today’s agreement illustrates our belief that copyright is a cornerstone of the film and television industry,” MPA Chairman and CEO Charles Rivkin said in a statement, adding that the deal “reflects our shared determination to continue our work together to further fortify those guardrails.” Bytedance general counsel, John Rogovin, said the company “respects the intellectual property rights that underpin creative industries around the world” and said the MOU was “an important framework for continued collaboration as the technology evolves.”
OpenAI’s Sora shutdown resonates with IP crackdown
ByteDance is not the first AI company the MPA has pressured into concessions on their AI product. The association had initially gone after OpenAI’s Sora before the firm added limits, according to Variety, with OpenAI eventually shutting Sora down earlier this year after a similar wave of unauthorized character videos.
However, instead of fully retreating and shutting down its products, ByteDance is taking a different path, pouring more funds into Seedance, which is reportedly one of the most advanced video models available, alongside Google’s offerings.
The Los Angeles Times reported that Seedance has gained ground with independent filmmakers who find it cheaper than rival tools, a development that gives ByteDance a commercial reason to keep the studios afloat instead of completely disengaging from the market.
If you're reading this, you’re already ahead. Stay there with our newsletter.

Groq has finalized a $350 million funding round valuing the AI chip startup turned data center operator at $3.5 billion, which equals almost half of the $6.9 billion the startup was worth only last September before Nvidia licensed its technology and hired away its founder.
Groq fall began from Nvidia poach
Groq started in 2016, building its own hardware to take on Nvidia in AI processing. This changed after Nvidia struck a licensing deal for the AI company’s technology and poached CEO Jonathan Ross along with other senior staff.
A deal of this type, alongside the subsequent change in staff has become prevalent in the AI industry. Google and Meta have done similar, licensing technology and bringing in engineers from AI startups instead of acquiring the companies. Large tech firms do this to secure great expertise without buying the business entirely.
The Nvidia deal stripped out leadership at Groq, leaving a company dedicated to rebuilding its staff and purpose over the last months.
$350 million round led by Disruptive
The round was led by Disruptive, a Dallas investment firm, according to Bloomberg. A Groq representative also said that Nvidia was investing money into the deal.
Disruptive was founded by Alex Davis, who also serves as Groq’s executive chairman.
The new valuation hits at about half of where the AI startup sat at its peak value. This September peak was stated to be $6.9 billion, pointing at a sharp reset for a company that was, until recently, one of the bigger names in the race to build silicon chips that could compete with Nvidia’s.
From selling chips hardware to renting compute
Groq has now restructured itself as a data center operator, chasing demand for the computing power that runs AI models in production, an area known as inference. The company raised $650 million in June to fund this shift, and reset its valuation at the time without attributing any specific numbers.
“Inference will without a doubt become the largest and most critical layer of AI infrastructure,” Davis said in a statement to Bloomberg News, adding that the startup would concentrate on “supporting the most important model makers.”
Some of the fresh capital from the round has been set aside for the startup’s physical expansion. Groq plans to grow its total data center capacity to more than 200 megawatts by next year.
The smartest crypto minds already read our newsletter. Want in? Join them.

Crypto company backed by President Donald Trump, World Liberty Financial, is connected to WorldClaw, a Hong Kong platform that offers access to Chinese AI models that have been marked by the U.S. government as security risks, according to a report by Reuters on Monday.
Flagged Chinese labs fill WorldClaw AI menu
WorldClaw has a catalogue of about 90 AI models on its routing service platform, with a Reuters review stating that 43 of them come from Chinese developers, including Alibaba, Baidu, Z.ai, DeepSeek, and Moonshot. OpenAI and Anthropic also have models available on the platform.
Several of these Chinese firms have particular U.S. government labels regarding their use. The Department of Defense has named Alibaba and Baidu as Chinese military-affiliated companies, a status barring the Pentagon from working with them in any capacity.
Z.ai, formerly Zhipu AI, sits on the Commerce Department’s entity list, which restricts its access to U.S. technology and forces American firms to seek an export license that is presumptively denied. The government listing also accuses Z.ai of advancing China’s military capabilities through advanced AI.
DeepSeek and Moonshot complete the names of marked out AI companies. Trump admin officials have accused both of building their systems on intellectual property stolen from American rivals. Moonshot has denied these allegations, however.
USD1 stablecoin is the connection
WorldClaw accepts USD1, World Liberty’s dollar-pegged stablecoin, as a form of payment for its services on the platform. USD1 is backed by multiple assets including U.S. Treasury securities, and the Trump family is entitled to a cut of the interest earned from these reserves. Every purchase completed using the USD1 stablecoin helps towards sustaining the WLFI venture partly owned by the Trump family.
Token sales at World Liberty have delivered the Trump family more than $1.4 billion, part of The Trump family has acquired about $2.3 billion from crypto in total. The firm recently won preliminary U.S. approval for a national bank charter built around the same USD1 stablecoin.
It is also worth noting that Ryan Fang, World Liberty’s head of growth, has previously advised WorldClaw in a role the AI platform has described as strictly advisory and focused on USD1 adoption and wider access to AI services. Donald Trump Jr. and Eric Trump have also previously promoted WorldClaw on X.
Risks are even greater than trade policy
A senior fellow at the Center for New American Security who previously advised the State Department, Daniel Remler, has warned that Chinese models available on WorldClaw could expose users to Chinese government monitoring and malicious code that could potentially hijack independent AI agents.
WorldClaw also states on its website that the information users enter may be passed on to the companies whose models power the platform.
The WorldClaw platform claims it has over 10,000 users and handles over 50 million requested tasks a day, a demand driven in part by the lower cost of Chinese models. Usage of these models remains generally legal for individuals and companies in the U.S..
Both sides claim no “conflict of interest”
Both companies have vehemently rejected the consensus that this connection is a conflict of interest. David Wachsman, a World Liberty spokesman, said WorldClaw was an independent company and stated that large U.S. firms also resell Chinese-built AI. A WorldClaw spokesperson has claimed that the platform “helps American AI companies reach international users” and that the mdoel listings on the platform do not “constitute an endorsement of its developer.”
The White House offered the same defense. Spokesperson Anna Kelly said there “are no conflicts of interest” in the relationship and that “President Trump only acts in the best interests of the American public.”
However, external experts remain not as convinced. “As the U.S. government tries to respond to the rise and threat of Chinese AI, it seems hypocritical to go out through WorldClaw to use these tools from China to try and make a bunch of money,” said Sam Bresnick, a fellow at Georgetown University’s Center for Security and Emerging Technology.


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

Harmony has published a detailed rollback plan that would rewind its blockchain to cancel the effects of an attacker who forged billions of ONE tokens.
The move will discard every transaction confirmed on the network since the incident, deleting legitimate user activity along with the fraud.
What is Harmony doing to its blockchain?
The Harmony network has announced that it will return to the state it was in on August 11, 2026, at 23:25:37 UTC. All transactions that happened after that date and time, including the fake ONE tokens that were created by a hacker, as well as any legal buys, sells, or transfers that users made, will be erased.
The company chose this specific time because the first fake tokens appeared in the very next block on Shard 0, which was block number 92,730,036.
The block just before it, number 92,730,034 (Shard 1) was never attacked, but it is being rolled back at the same time to prevent any future problems. The network will then start creating new blocks from heights 92,730,035 and 94,978,279.
Harmony’s new client software, v2026.1.2, will refuse to accept the block hashes tied to the hack.
Before the team decided on a fixed rollback window, a targeted burn to destroy the fake token was considered as an initial solution. However, the fake ONE tokens had already been sent to exchanges, trading pools, and smart contracts; burning them could have accidentally taken money that belongs to innocent people.
The company also rejected the idea of a blacklist, because that would leave the fake supply in the system, and selectively replaying transactions was called “unworkable” because the state of the chain has changed. The idea of a full token migration was also rejected because it would have caused too much disruption for all users.
Simply using a built-in revert tool to move the chain head was also warned against, as this could leave behind unwanted data that might lead to another failure.
Harmony Network’s report states that an independent security firm reviewed the hack and agreed with the rollback solution.
How did the hack happen?
The hack was first noticed by on-chain expert “Juiceberg,” who estimated that the attacker used a flaw in how the network checks cross-shard receipts to process the same receipt more than once, creating tokens out of thin air.
After the attack was made public, the price of the ONE token fell, and it is currently around $0.0007.
Roughly 4 billion ONE, about 26% of the token’s supply, had been conjured through empty blocks.
Harmony says it is now investigating where the money went. The hacker tried to send 534 payments of 5 billion tokens each in just 106 seconds. Of those, 477 payments went through, moving a huge amount of 2,385,000,000,000 ONE.
The team tracked the flow of money and found that most of the fake tokens ended up in a wallet or reached a service like an exchange. Harmony says it is now working with exchanges, crypto bridges, and police to keep records and find the criminals.
Cryptopolitan reported that Harmony had a similar problem in December 2023, when a bug in its staking system created 146.28 million ONE tokens by accident, forcing the company to update its network.
Separately, Harmony’s Horizon bridge lost about $100 million in a June 2022 theft that the FBI attributed to North Korea’s Lazarus Group.
The smartest crypto minds already read our newsletter. Want in? Join them.