oAdam

Is it definitely safe to factory-reset an old iPhone, turn on Airplane Mode, and use it as an offline seed phrase vault?

Yishi, founder of OneKey, warns that if the device is an iPhone 11 or earlier, the answer may not be so reassuring.

The risks stem from BootROM vulnerabilities deep in the chip, including checkm8 and usbliter8. These affect models such as the iPhone XS, XR, and 11 series, as well as the second-generation SE.

If an attacker gains physical access to the device, they can connect a specially crafted USB device in DFU mode and execute code before the operating system starts.

These attacks pose several challenges:

* Airplane Mode cannot stop them, because the attack happens before the system starts.
* A factory reset and software updates cannot fix underlying hardware vulnerabilities.
* After an attack, the phone may show no obvious signs of tampering in its appearance or everyday operation.

This is why so-called “evil maid attacks” are a concern: an attacker doesn’t need long-term control of the device. A single opportunity for physical access may be enough to break the security assumptions you’ve been relying on.

Yishi’s recommendations include:

1. Check your iPhone model. If it’s one of the older models mentioned above, migrate your seed phrase as soon as possible.
2. Generate a completely new seed phrase and transfer your assets to a new wallet. Simply deriving new addresses from the old seed phrase does not eliminate the risk posed by the old seed.
3. Use a sufficiently long alphanumeric passcode; don’t rely solely on a simple six-digit PIN.
4. Bring your own charger and avoid using USB devices from unknown sources.
5. Turn on iPhone Lockdown Mode and keep the system up to date.

Many people equate an offline device with a secure device. But the true security boundary isn’t just about network connectivity—it also includes hardware, the boot chain, and physical access.

Keeping your seed phrase safe requires more than simply disconnecting from the internet.

China’s AI competition is no longer just about model benchmark scores and parameter counts. The real changes are happening in everyday use.

As of June 2026, China’s daily AI usage has exceeded 175 trillion tokens, an increase of more than 1,000 times compared with early 2024. The latest report from the China Center for Information Industry Development (CCID) notes that, for the first time, China’s AI inference data volume has surpassed its training data volume—a sign that the industry is shifting from model training and development to large-scale practical applications.

Model iterations are also accelerating. The update cycle for leading large language models has shrunk from months to weeks, or even days.

This change is making its way into manufacturing:

* More than 500 advanced smart factories have been built nationwide, spanning all 31 provincial-level regions and over 80% of manufacturing industry categories.
* Domestic computing power and large language models have been deployed in over 1,000 application scenarios across 20 industries, including steel and automotive.
* New manufacturing models, such as personalized customization and full-lifecycle services, are beginning to take shape.

The user base is growing too. As of the first half of 2026, China had more than 700 million generative AI users, with a penetration rate exceeding 50%.

The significance of 175 trillion tokens lies in more than just a huge usage figure. It reflects how AI is evolving from a technology tool used by a small number of people into a foundational capability embedded in business processes and everyday life.

In the next phase, the focus of competition will become increasingly clear: those who can truly integrate AI into business processes will be better positioned to turn technological advantages into productivity.

Websites finally have a way to tell the entire AI industry, with a single rule: content here may not be used to train models.

OpenAI, Google, Microsoft, and Anthropic have announced support for a new universal standard that lets websites use robots.txt to declare in one place whether AI companies may crawl their content for model training—without having to configure settings separately for each company.

The standard was submitted to the IETF by Anthropic technical lead Kevin Kelley and is integrated directly into websites’ existing robots.txt files.

Until now, websites have faced a patchwork of independent rules for AI crawlers. To prevent one company from crawling content for training, a site had to configure a separate rule; to deal with another company, it had to make another declaration. The more AI companies there are, the harder it becomes for websites to keep up.

The new standard aims to solve this problem by giving content publishers a unified, machine-readable signal, which AI companies can use to identify websites’ preferences about training permissions.

Companies are at different stages of implementation:

* Anthropic has already adopted it.
* OpenAI plans to implement it in the coming months.
* Google and Microsoft are expected to support it next year.
* Meta says it supports establishing a universal signal, but has not announced a specific implementation timeline.

The standard does not mean that websites can now completely prevent all AI crawling. Its practical effect still depends on whether crawlers follow the rules.

But the direction is clear: AI companies need internet content, and content publishers should have the right to express their consent preferences. In the past, each platform had its own set of rules; in the future, they may be replaced by a single standard adopted across the industry.

In the new week, markets will have to digest three things at once: a pullback in precious metals, U.S. economic data, and earnings from AI giants.

First, a look at last week's performance:

* Spot gold closed at $3,876.89 per ounce, down 6.7% for October.
* Spot silver closed at $49.13 per ounce, down more than 18% in October.

Precious metals have undergone a marked correction. Markets will next face tests from expectations for Fed policy, U.S. employment data, and inflation indicators.

Key events next week (Beijing time)

Tuesday

* 2026 U.S. midterm elections; exact timing to be announced.

Wednesday

* 18:00: Eurozone October CPI, preliminary year-over-year and month-over-month readings.
* 21:30: U.S. September trade balance; U.S. Treasury quarterly refunding announcement.
* 22:45: U.S. October S&P Global Services PMI, final reading.
* 23:00: U.S. October ISM Services PMI.

Thursday

* 18:00: Eurozone September PPI, month-over-month.
* 21:30: U.S. initial jobless claims for the week ending October 31.

Friday

* 17:30: New York Fed President Williams, ECB Chief Economist Lane, and others take part in a discussion at the ECB's money market conference.
* 21:30: U.S. October unemployment rate, nonfarm payrolls, and average hourly earnings, year-over-year and month-over-month.
* 23:00: U.S. University of Michigan November consumer sentiment index, preliminary reading, and preliminary one-year inflation expectations.

The earnings calendar is also packed.

More than 100 S&P 500 constituents will report results next week. LSEG IBES data show that more than half of companies have reported third-quarter results, with S&P 500 third-quarter earnings growth expected to reach 31.6%.

Apple will report earnings after the market closes on Monday, U.S. Eastern Time. AMD and Qualcomm will also report results next week.

There are plenty of market-moving factors this week: employment and inflation data will shape interest-rate expectations, tech giants' earnings will test the returns on AI investment, and precious metals will have to contend with market sentiment following their recent correction.

In the coming days, macroeconomic data and tech stock earnings could both become catalysts for market volatility.

China’s AI and chip industries are showing three different growth curves: some are pushing into advanced manufacturing, some are investing in R&D for the future, and others are beginning to grapple with the operational strain of a product taking off.

ChangXin Memory Technologies’ HBM3E samples have reached domestic chip design companies. Although yields are currently only 20%–30%, the company still plans to begin limited mass production within two months.

Its conventional memory business is enjoying another wave of growth: third-quarter revenue was 11.871 billion yuan and net profit was 5.44 billion yuan, up 613.2% year over year but down 19.3% from the previous quarter. As global memory makers shift more capacity toward higher-value products, they are also creating room for conventional memory prices to rise.

Unitree presents a different picture. Third-quarter revenue rose 69.9% to 665 million yuan, while attributable net profit fell 17.4% to 112.7 million yuan. R&D spending increased by about 250%, and the gross margin dropped to 50.6%. Revenue is still growing, but profits are under pressure, as R&D investment is clearly accelerating.

Moonshot AI’s Kimi K3.1 Pro has expanded its context window to 1 million tokens and promised to release the full model weights on November 4. However, shortly after launch, users encountered rate limits and unexpected usage of additional credits. The team later apologized and restored a quarter of the monthly quota to subscribers.

The three companies are taking different paths, but they all face the same reality: technological breakthroughs, revenue growth, and profitability are never the same thing.

The money North Korean hackers have stolen from the crypto industry is now equivalent to nearly half of the country’s GDP.

On-chain analyst Darkfost estimates that the Lazarus Group has stolen a total of $11.87 billion worth of crypto assets.

Compared with North Korea’s economic output in 2025, that figure is equivalent to 45% of its GDP and 35% of its gross national income.

In other words, the assets a single hacking group has taken from the global crypto market through cyberattacks now amount to a staggering share of a country’s total economy.

This also explains why the Lazarus Group remains one of the threat groups the crypto industry needs to watch most closely. For the group, attacking crypto platforms is not just cybercrime—it could also be an important source of foreign currency.

It should be noted that the $11.87 billion figure is Darkfost’s cumulative estimate, not North Korean revenue confirmed by an official audit.

AI-generated text is starting to develop its own “invisible fingerprint.”

Starting November 7, SpaceXAI will enable text watermarking globally for several models in the Grok API, including Grok 4.7, 4.6, 4.5, and 4.3, as well as three versions of Grok 4.20: Reasoning, Non-Reasoning, and Multi-Agent.

The mechanism doesn’t rely on hidden characters or embedding information in text metadata. Instead, it adjusts token selection during generation, weaving the watermark signal into the text itself. It’s imperceptible to the human eye, but specialized detectors can identify it.

SpaceXAI said the move is in response to the EU’s Artificial Intelligence Act. Team administrators can also disable watermarking for specific models.

Andrew Kurlan noted that ChatGPT, in North America, is currently the only model that still doesn’t add invisible watermarks to generated text.

AI-generated content is entering a new phase: until now, users have mainly had to judge whether text sounds like it was written by AI. In the future, technical signals may make it possible to trace where it came from.

Of course, a watermark can offer clues about a text’s origin, but it cannot, on its own, prove who authored it.

Premia Labs on Arbitrum is under an ongoing attack, and approximately $400,500 in treasury assets has been transferred.

Security firm Blockaid detected that the attacker exploited a vulnerability in the mechanics between the options pool and the treasury.

The operation can be broadly divided into three steps:

1. Create an empty options pool.
2. Use a flash loan to buy options, locking the treasury’s collateral in the relevant pool.
3. Then exploit the liquidity mechanism to withdraw treasury assets.

By the time Blockaid released its report, the attacker had transferred approximately $400,500 in about two transactions, and the attack was still ongoing.

What makes this attack concerning is that the attacker exploited the interactions between different mechanisms within the protocol, rather than simply stealing users’ private keys.

Official confirmation from Premia Labs is still needed regarding the cause of the vulnerability, the scope of its impact, and whether the relevant functionality has been paused.

The $400,500 is the amount transferred so far that has been disclosed, not the final loss. The attack is ongoing, so subsequent fund movements and the protocol’s response will be more important.

Troves experienced a security incident, with the xWBTC/WBTC vault initially losing approximately 8.6 BTC.

The attack did not target the new contracts currently in use. Instead, it exploited an older asset rebalancing contract that had been replaced but still retained operational permissions.

Troves stated that the relevant permissions have been revoked.

To prevent the risk from spreading, the platform paused three vaults:

* xWBTC/WBTC
* xSTRK/STRK
* xstrkBTC/strkBTC

So far, no actual losses have been found in any other vaults.

The platform is working with security firms to track the stolen funds and explore recovery options. It will later publish a detailed incident analysis and arrangements for affected depositors.

This incident once again highlights a potential vulnerability in DeFi contract management:

Launching a new contract does not mean the permissions of the old contract are automatically revoked.

If an old version retains operational permissions, it can become an entry point for attackers—even if it is no longer responsible for core operations.

The next things to watch are how much funding Troves can recover and the specific compensation plan for affected depositors.

Crypto primary market fundraising surged 83.5% month over month in October, but one figure tells the story:

Kalshi alone took 51.2% of all funding raised that month.

According to RootData, 53 crypto primary market funding rounds were announced in October, totaling approximately $1.954 billion.

That was up 83.5% from $1.065 billion in September, but still down 45.3% from $3.573 billion in October last year.

The number of funding rounds was also just 53, below the 78 recorded during the same period last year.

Funding was concentrated in three sectors:

* DeFi: 17 rounds, approximately $1.157 billion
* Infrastructure: 17 rounds, approximately $348 million
* CeFi: 9 rounds, approximately $343 million

The three largest individual funding rounds were:

* Kalshi: $1 billion
* Whop: $200 million
* Pearl: $130 million

Nous Research and Spiko each raised $90 million.

The five largest projects raised a combined $1.510 billion, accounting for 77.3% of the month's total. This means that, excluding a handful of large deals, other projects raised considerably less.

In the RWA sector, six related projects announced funding in October, totaling approximately $122 million. Spiko completed a $90 million Series B round, while Fission and Multiliquid raised $10 million and $8 million, respectively.

The M&A market recorded 11 deals, unchanged from September, but deal values were not disclosed.

Among investors, Tether participated in four funding rounds; Sequoia Capital, Pantera Capital, and North Island Ventures each participated in three.

These figures make me more inclined to describe October as a funding rebound driven by leading projects, rather than a broad-based recovery across the entire crypto primary market.

Total funding did rise, but the figures for the same period last year were still higher, and capital was clearly concentrated in the hands of a few projects.

Money in the market doesn't mean every sector has money; and leading projects raising huge sums doesn't mean funding conditions have improved for ordinary startup teams.

Tether earned another $1.5 billion in the third quarter, bringing its cumulative net operating profit for the first three quarters to $4.5 billion.

This is no longer just the revenue scale of a stablecoin issuer. It is a business that continuously generates returns from a huge reserve asset base.

A few figures from the latest financial report are worth remembering:

* Q3 net operating profit: $1.5 billion, unchanged from the previous quarter
* Cumulative profit for the first three quarters of 2026: $4.5 billion
* USDT in circulation: approximately $184 billion
* Reserve buffer: $7.95 billion, nearly double the $4.1 billion at the end of Q2

As of September 30, Tether had approximately $191.5 billion in total reserve assets and around $183.5 billion in liabilities, leaving assets in excess of liabilities by about $7.95 billion.

Its direct and indirect holdings of U.S. Treasuries reached $132.4 billion. U.S. Treasuries and reverse repurchase agreements remain Tether’s primary sources of income.

This reveals how Tether makes money:

Users hold USDT, while Tether manages the corresponding reserve assets and earns returns through financial instruments such as Treasuries. The larger the amount in circulation, the greater the pool of assets it can manage.

But profits aren’t the only thing that matters for a stablecoin issuer. The real test of this business is whether its reserve assets are sufficient, whether their liquidity is reliable, and whether it can meet redemptions when many users seek to redeem at once.

$4.5 billion in profit for the first three quarters, and a reserve buffer of nearly $8 billion. Tether is no longer just a stablecoin issuer in the crypto market; it is a financial player with a massive balance sheet.

b14g experienced a security incident, in which approximately 97 million CORE were withdrawn without authorization.

At 09:29 UTC on October 31, an attacker transferred approximately 97 million CORE from b14g’s staking contract to an address under the attacker’s control.

After discovering the anomaly, the project team paused the affected contract.

Current fund-tracking progress:

* Approximately 68 million CORE remain on the Core network
* Exchange partners have frozen approximately 9 million related CORE
* BTC staked through b14g remains time-locked on the Bitcoin network and has not been affected

b14g is working with the Core team, security partners, and exchanges to track the funds and coordinate efforts to freeze and recover them.

The project team said its top priority is protecting users and recovering the assets. A full post-incident report will be published later, explaining the technical cause, the actual scope of the impact, and arrangements for affected users.

The two attacker addresses disclosed in this announcement:

0xc556ed208285406d71705927f92d4f4eaea579f1

0x27758c30f8e424fbd8ca605bf31e9496c26ce92c

Please note that the incident affected CORE staked through b14g, not all staked assets. BTC remains locked on the Bitcoin network and has not been affected.

The transfer of 97 million CORE is only the beginning of the incident. The most critical questions are how much of the funds can be recovered and how users will be compensated.

The replenishment rate for assets stolen from Binance has risen to 90%, with ten major assets now fully covered.

Binance’s latest security incident update shows that more than $6.34 billion in funds set aside earlier is being used to replenish the stolen assets. The overall replenishment rate has now increased from about 80% to 90%.

The ten assets that have been fully covered are:

BTC, ETH, USDT, BNB, SOL, USDC, XRP, USD1, RLUSD, ZEC.

The remaining approximately 10% consists of assets outside these ten. They are relatively less liquid and will take more time to acquire through market and over-the-counter channels.

The replenishment funds come from three sources:

* Some of the funds in the SAFU user protection fund, expected to be deployed within the next 24 hours
* Some of the funds in the futures insurance fund
* Binance’s own corporate funds

Binance has also committed to restoring the SAFU fund to approximately $1 billion.

Regarding the proof of reserves, Binance completed a new asset snapshot on October 31 at 08:00 (UTC+8) and updated its proof-of-reserves page. Generating the full zero-knowledge proof is expected to take 40 to 50 hours, with completion expected on November 3.

Withdrawals are still proceeding as planned, with a target reopening date of November 1. Further updates will be announced later.

One point Binance has repeatedly emphasized throughout this incident is that the platform will bear the financial losses caused by the stolen funds, and users will not be held responsible.

90% has been replenished, and the rest is still being acquired. What remains to be verified is whether all the remaining assets can be replenished, and whether withdrawals and proof of reserves can resume as planned.

If AI meant you no longer had to work and gave you 100% of your current salary, would you quit?

A survey of 8,234 working-age adults in the UK, the US, Denmark, and India found some interesting results:

75% of respondents said they would be willing to leave their jobs if they could receive a guaranteed income equal to 100% of their current salary.

But another group gave the opposite answer:

20% said they would not quit even if they received 120% of their current salary.

And this proportion was about the same across different income groups.

More thought-provoking still, only 26% of people ranked pay as the most important thing they valued in a job, while about a third felt that the work itself mattered more.

This means that salary alone cannot explain why people are willing to walk into the office every day.

Work also provides a sense of purpose, social connections, identity, and the feeling of being needed. For some people, losing these things cannot be made up for simply by getting more money.

Paper author Julian Jacobs argues that policies during the AI transition should focus not only on income security, but also on protecting the sense of purpose, community, and dignity that work provides.

AI may change how people work, and even make some jobs unnecessary.

But one question cannot be ignored:

When someone no longer needs to earn a living through work, where else can they find meaning in life?

You might think you’ve just given an AI Agent a new skill. In reality, you may also have inadvertently removed human approval.

A study of public GitHub repositories scanned around 1.53 million English-language Agent skills and found that about 88,000 contained instructions that could bypass human review.

That works out to roughly 1 in every 17 skills containing this kind of instruction.

Most concerning of all:

70.1% of these instructions don’t distinguish between usage scenarios.

For example:

* Execute and push changes without confirmation.
* Don’t ask questions along the way; keep going until the task is complete.
* Handle every step automatically, without human intervention.

If instructions like these are written into a SKILL.md file, an Agent loading that skill could act directly at steps that would normally require human confirmation.

Even more troubling, existing security scanners may not catch them.

The study tested two open-source scanners from Cisco and NVIDIA. Even when using GPT-5.4, their highest detection rate was only 46%, with precision below 0.4.

But change the way you ask the question, and the results are markedly different:

Instead of asking the model, “Is this a malicious instruction?”, ask, “Does it bypass human review?” The detection rate rises to 61%, and precision reaches 0.71.

This points to a problem:

Agent security reviews can’t focus only on malicious code. They also need to check instructions that may appear to be merely about improving automation efficiency.

The study’s recommendations are straightforward:

1. Set clear limits on instructions that trigger automatic execution—for example, allow them to take effect only in a CI environment.
2. When reviewing skills, focus on whether they bypass human confirmation, not just on scanning for malicious code.
3. Don’t let a single unconditional instruction determine an Agent’s behavior in every scenario.

Agents are becoming more capable—and gaining more and more permissions.

But some steps shouldn’t stop requiring human confirmation just because someone wrote, “Keep going until it’s done.”

Truly reliable automation doesn’t eliminate every check. It knows when it needs to stop and ask a person.

Tech giants are starting to answer for their AI investments.

In 2026, five companies—Meta, Alphabet, Oracle, Amazon, and Microsoft—have collectively pledged to invest $800.6 billion.

This isn’t an abstract vision of AI. It’s real spending on data centers, servers, networking equipment, and infrastructure.

The market is taking a fresh look at this arms race.

Alphabet’s AI Overviews are beginning to affect its core search advertising revenue, while the company’s capital spending this year is expected to reach as much as $205 billion. AI is changing search—and could also disrupt Google’s most important way of making money.

Intel’s CFO said this year’s cash inflows are expected to fall short of spending, and that weakness in the PC market will persist at least through the first half of next year. The company is also continuing to ramp up its investment in equipment.

Amazon has pledged to invest $220 billion in AI infrastructure this year, with cash spending also exceeding revenue. After its earnings fell short of expectations, its share price briefly dropped 6% in after-hours trading.

Meanwhile, Reddit fell 26% as daily active users missed expectations and Google’s AI-powered search threatened referral traffic. Roblox shares dropped nearly 20% as daily active users declined.

Intel plunged as much as 30% intraday, Alphabet lost more than $500 billion in market value the previous day, and the Nasdaq fell more than 3.5% on Friday.

This wave of declines sends a signal:

The capital markets haven’t suddenly stopped believing in AI. They’re starting to demand that AI companies prove their investments can generate returns.

In the past, growth projections alone could command a premium. Now, investors are watching cash flow, capital expenditures, and progress toward commercialization.

AI remains one of the technology industry’s most important investment areas. But no matter how compelling the story, someone has to pay the bill.

Finding a restaurant in the future might not even require opening an app.

Zest has launched a personal AI restaurant concierge. Send a text, and it can find restaurants, compare options, organize a group vote, and complete the reservation.

It has a particularly interesting feature: recommendations are based not only on preferences users enter themselves, but also on their actual dining spending history.

Zest imports credit card statements through Plaid and analyzes only dining transactions to build a taste profile for each user. The places you’ve often visited and the kinds of restaurants you enjoy become reference points for the AI’s recommendations.

No more endless back-and-forth in the group chat when planning a meal with friends.

Add Zest to the chat, and it can take everyone’s tastes into account, suggest restaurants, run a vote to choose a spot, and then book a table.

Reservations are handled through OpenTable, Resy, and SevenRooms. You can search by cuisine, neighborhood, occasion, and dietary restrictions, as well as ask whether a restaurant is crowded, whether it has Wi-Fi, or even request its menu directly.

Founded in November 2024, Zest raised $1.8 million in pre-seed funding from 776 and Kindred Ventures.

The founder’s product vision is also clear: the AI agent itself is the product. Reservations and food ordering are the initial paths to monetization, with merchant recommendations and shopping as possible future extensions.

What I find interesting about this approach is that it doesn’t turn AI into yet another chat window.

Users don’t need to figure out how to phrase their questions or switch back and forth between multiple platforms. Just explain what you need, and let the AI handle the search, coordination, and booking.

A truly useful AI assistant doesn’t just give you answers—it gets things done.

Phone number: (212) 470-6067

McDonald’s, Disney, and PayPal stock tokens are starting to be used to play with meme coins.

A new way of combining assets is emerging on Robinhood Chain: stock tokens are no longer just investments for users to buy and sell—they can also serve as trading pairs and liquidity reserves for meme coins.

Pons Treasury is driving this model.

It has already provided liquidity for tokenized $MCD, $DIS, and $PYPL, making them available for trading on Robinhood Chain. The latest stock tokens that can be paired include $BMNR, $JPM, $V, $UBER, and $GDX.

More interestingly, Pons wants to do more than just facilitate stock token trading—it wants users to launch and trade new meme coins directly using these tokens.

This means stock tokens could evolve from standalone investment assets into liquidity tools within the meme ecosystem.

Until now, meme coin trading has largely revolved around stablecoins and crypto-native assets. Now, traditional stock tokens could also become part of the mix.

If this approach scales, the range of assets and trading combinations in the on-chain meme market could become much richer.

Of course, stock tokens are not the same as directly owning the underlying companies’ shares. The specific rights, redemption mechanisms, and liquidity still depend on how each product is designed.

When stock tokens for McDonald’s and Disney can serve as trading fuel for meme coins, the gap between traditional finance and on-chain speculation narrows another step.

Google is entering the crypto custody market, targeting banks and large institutions.

For institutions entering crypto, there has always been one hurdle they can’t avoid: private key security.

Who holds the assets, who has the authority to sign, how internal approvals work, and what happens if keys are stolen—these issues can’t be solved by simply giving an institution a regular wallet.

Google’s solution is an MPC key management and signing service.

It splits private keys into multiple key shares, which are stored across Google-owned hardware, Google Cloud, and other cloud environments. When a signature is needed, multiple shares work together to complete it, so the full private key doesn’t need to be kept in one place.

Institutions can also configure rules such as transaction limits, whitelisted addresses, and approval by multiple departments, extending their existing internal risk controls to on-chain transactions.

Most importantly, institutions retain control of the key shares, and Google cannot directly access the complete private key.

The product is currently in the Private Preview stage, with several large financial institutions testing it. Alphabet is also using it to protect its digital assets on Ethereum, and plans to make it commercially available in early 2027.

This reflects a broader trend:

Crypto is gradually shifting from a financial experiment for individual users to infrastructure that traditional financial institutions need to connect to.

Banks don’t necessarily need to build custody systems from scratch, but they do need solutions that meet their own security standards and requirements for access management and compliance.

That’s the layer of infrastructure Google aims to provide.

If banks can use mature MPC technology to manage on-chain assets in the future, the technical barriers to institutional adoption of crypto could fall further.

What’s truly exciting isn’t just the arrival of another custody product, but the beginning of traditional financial infrastructure extending on-chain.

AI is changing how the rewards from crypto airdrops are distributed.

In the past, studios made money because they had technical skills that ordinary people didn't: running interactions at scale, writing automated scripts, analyzing on-chain data, and managing accounts.

Projects needed these capabilities to generate user numbers, trading data, and market buzz, and paid for them with the promise of future tokens.

But now, AI has lowered the technical barrier.

An ordinary person can also use AI to build scripts, bots, and data tools. Studios no longer have the clear informational and technical edge they once did.

The problem is that projects have also benefited from the cost reductions brought by AI.

Tasks that once had to be outsourced to studios can now be handled in-house; on-chain data that once had to be incentivized through airdrops is no longer so scarce.

Naturally, projects will reconsider how to allocate their token budgets: should they go to addresses farming interactions, or be invested in liquidity, exchanges, and channels that can actually generate revenue?

That's why I believe traditional airdrops are becoming less attractive.

In the future, it may be more worthwhile to study early-stage projects with real profitability, as well as products that distribute tokens through public sales, user benefits, and other means.

Another direction is projects that are part meme, part airdrop farming, where users' trading, promotion, and community activity are themselves part of the product ecosystem.

But with perps and prediction markets, you have to calculate the costs carefully. Fees, slippage, and trading losses all cost real money. Points aren't the same as returns, and trading volume isn't the same as profit.

Now that AI has leveled the technical playing field for airdrop farming, the real competition is no longer about who can farm the most, but who can spot worthwhile projects early.