周东野

OMG, my God!! The AI world is about to change!!

The U.S. Department of Justice (DOJ) has announced the creation of a “Superintelligence Task Force.”
On the surface, it aims to crack down on foreign hackers, criminals, and fraudsters abusing advanced AI. But what’s really worth paying attention to is just how far its regulatory reach extends—requiring AI companies to strengthen “internal controls” and ensure their models “behave as intended.”

“Behave as intended” sounds like a safety measure, but at its core, it puts “compliance shackles” on frontier models.

For frontier AI, what counts as “intended behavior”?
Who gets to define that “intention”?

When regulators begin intervening in the behavioral boundaries of models, it means safety oversight of large models has moved beyond simply “preventing cybercrime” and into the realm of “compliance and control.”

In this policy environment, the regulatory siphon effect in the frontier AI sector will only grow stronger:

Consolidation among leading models and soaring compliance costs: Only deep-pocketed giants like OpenAI, Anthropic, and Google can afford the large compliance teams needed to handle DOJ safety audits.

Smaller and midsize innovation teams will find themselves increasingly disadvantaged by compliance requirements.

Open-source AI will face even tighter constraints: When “preventing model misuse” becomes the highest administrative priority, open-source models that can be freely deployed and fine-tuned will be first in the firing line.

The policy divide between open-source and closed-source AI may continue to widen.

The underlying rationale for decentralized AI (DeAI) is indirectly validated: When control over compute, models, and data is highly concentrated in the hands of a few strictly regulated entities, market demand for permissionless compute networks, censorship-resistant model hosting, and privacy-preserving data collaboration will become even more apparent.

This move by the U.S. Department of Justice marks a new phase in AI regulation, shifting from “being wary of technological risks” to “intervening in model governance.”

What we need to keep watching is not how many more AI criminals they arrest, but whether this control will directly affect the room for growth of the open-source ecosystem and Decentralized AI.

#Aİ #美股遭遇2020年以来最严重抛售

What the hell is that garbage DK doing? Engaging before everyone's here, engaging with no abilities available—it looks like a complete newbie's playstyle.#币安神算子

Polygon’s Hard Fork Dilemma: Recover Stolen Assets or Uphold Its Principles?

Polygon is weighing whether to recover assets stolen by hackers through a hard fork—a decision that will test blockchain immutability and its reputation for decentralization. Technical analysts are watching this key level, and indicators suggest Polygon’s future depends on the community’s choice.#Polygon #BlockchainSecurity

De¹ Launches Ginza 1.0 Financial Model, Ushering in a New Era of On-Chain Prediction

De¹ has unveiled its latest financial decision-making model, Ginza 1.0. Its real-time forecasting capabilities deliver precise insights for on-chain trading, ushering in a new paradigm for intelligent financial prediction. Now available for a limited trial—don’t miss out.

Google tells US K–12 schools: If you want to turn off AI, you’ll have to turn off Search too.

Google announced this week that, starting next year, AI tools will be off by default on student accounts, with schools deciding whether to turn them on—but Search counts as AI too.

After getting criticized that day, Google clarified that schools can turn on Search by itself. But the AI summary at the top of search results still can’t be turned off, just like in the version we usually use.

There are about 50 million Chromebooks in US schools. 【The search box is AI’s most reliable entry point】, and Google won’t let schools turn it off.

DuckDuckGo’s CEO seized the opportunity to post a tutorial on X showing schools how to change their default search engine.

Should AI summaries have an off switch, or is it fine to leave them on by default?

#Google $GOOGL

What does everyone think of the Pistons and 76ers today? I analyzed it all night, and based on the current score, I’ve decided to go all in on the Pistons. #币安神算子

Coinbase launches the x402 payment standard, ushering AI agents into the era of crypto payments

Coinbase’s open x402 payment standard gives AI agents a stablecoin payment solution. The use of USDC on the Base network marks a technological breakthrough in payments. More than 50 companies have joined, underscoring the milestone significance of this payment standard.

North Korean hackers' AI attacks target cryptocurrency and financial institutions

Microsoft's security blog revealed that North Korean hackers are using AI technology to comprehensively infiltrate target systems across multiple sectors, including cryptocurrency and financial institutions. Technical analysts advise watching key indicators and strictly adhering to risk limits.#朝鲜黑客 #AI攻击 #CryptocurrencySecurity

Everyone says, “If an exchange promises full compensation, users have nothing to worry about.” Brazil’s largest exchange, Mercado Bitcoin, is putting that claim to the test.

It was hacked on November 2. The exchange puts its losses at $36.3 million, while blockchain analytics firm Chainalysis estimates $49 million. The exchange says users won’t lose out and that it has recovered or frozen about $12.9 million.

But as of this morning, withdrawals are still disabled, with no date set for their return. The numbers are all there on your account, but you still can’t transfer your money out.

Compensation means “we’ll make it up to you later”; withdrawals mean “can you use your money now?” Those are two different things. $BTC fell back to 77,000 this week from 81,000, and people whose funds are frozen don’t even have the option to sell.

The most discussed item in today’s intelligence briefing was $HYPE . The breakdown is in the poster.

Tether’s CEO said at a Bitcoin conference in Lugano, Switzerland, that the company itself holds about 150,000 bitcoins, worth roughly $11.5 billion.

But its latest reserve report lists only the equivalent of about 100,000 bitcoins.

Tether is the company that issues the USDT stablecoin, and that report lists only “assets backing USDT.” The additional 50,000 bitcoins, worth about $3.8 billion, were not addressed by the CEO in terms of where they’re held or when they were bought.

Tether has previously said that its proprietary investments are funded by profits and kept separate from USDT reserves, so they don’t appear in the report.

Only one side can be verified: the 100,000 bitcoins are attested to by an accountant; the 150,000 figure is only a verbal claim.

Do you think Tether should disclose the addresses of those 50,000 bitcoins? Should / No need

I've already gone all in, IG and all—straight up 50k points. If I win, I'll make a run for the top 10; if I lose, I'm back to zero#币安神算子

ChatGPT Search Volume Nears Google’s as a New AI Search Landscape Emerges

OpenAI claims ChatGPT’s search volume has reached 20% of Google’s. AI search technology is developing rapidly and affecting the existing market landscape. Tech analysts are paying attention: key metrics show that ChatGPT is challenging Google’s dominance in search.

Binance targeted by hackers using fake job interviews to steal data; internal AI tools abused

Binance recently suffered a cyberattack in which hackers accessed employee accounts through fake job interviews and used internal AI tools to steal cryptocurrency. Technical analysts warn that security is the foundation of the crypto market, and that extra vigilance is needed at key levels.

Apple’s Q4 results just came out. On the surface, revenue hit a record $114.5 billion, up 12%. But strip away the distortions, and the underlying risks are huge.

This is really a set of books where a “one-time tariff refund” and “service revenue growth” are masking a hardware cost crisis.

Look past the gloss, and the key figures tell a tougher story:

Reported gross margin surged to 47.9%, which looks excellent. But that figure includes a full percentage point (nearly $1.1 billion) from a one-time tariff refund. Strip out that extra boost, and gross margin actually declined quarter over quarter. There’s only one reason: procurement costs for upstream memory chips (RAM/NAND) have surged at an unprecedented pace. Worse, management’s guidance says that the pass-through of higher memory chip prices, combined with the cost of cloud-based Siri AI inference, will drive next quarter’s gross margin down to 45%–46%.

The hardware business is sharply divided: • iPhone ($57.6 billion, +17%) and Mac ($10.8 billion, +24%) posted strong growth on new product shipments, but both face serious supply-chain capacity constraints. • iPad ($6.2 billion, -10%) and wearables ($8.3 billion, -8%) both stumbled. It’s not just a tough comparison with a high base: memory chip shortages have directly forced cuts to production capacity for lower-priced models.

Services ($31.6 billion, +10%) remains the business underpinning the company’s valuation. The number of active devices reached 2.5 billion, and paid subscriptions topped 1.5 billion. But note that guidance for services growth next quarter has slowed to 8%–9%. On top of that, Apple has now scrapped its “net cash neutral” financial target. If the underlying costs of a major expansion in hardware for Siri AI inference are pushed into services operating expenses, that means supply-chain costs are eating into hardware’s pricing power, while AI computing demands are eroding software margins.

On the surface, the results beat expectations on both revenue and earnings. In reality, this marks the beginning of a prolonged period of pain: expensive memory chips are squeezing margins, AI computing is eating into cash flow, and hardware capacity is stretched thin. With shortages compounding rising costs, the real test comes next quarter.$AAPL

#苹果财报喜忧参半

Anthropic included a chilling line in its prospectus: the company may soon build AI that can improve itself.

More subtly, the same filing shows revenue surging while its third-quarter net loss widened year over year.

In its S-1A filing, Anthropic warned that it may soon create models capable of recursive self-improvement (RSI), potentially accelerating AI capabilities sharply. It also said that human control would be maintained during research and development. The prospectus further noted that advances in model capabilities could outpace the development of safety measures.

Why does this matter?

The company is selling a story to the market while admitting in its risk disclosures that the story could spiral out of control. This is a public filing: errors carry legal liability. It’s not marketing spin.

Putting “AI building AI” in the risk factors shows this is no longer just a concept. But computing costs are eating into profits too: the faster it runs, the more it burns.

4 other stories worth a look today:

🔹 UK warns AI ransomware attacks on critical infrastructure are “highly likely” before 2028
→ Attacks are accelerating. Are your backups still stuck in the habits of three years ago?

🔹 North Korean hackers create a new wallet every 25 seconds to launder stolen Binance funds
→ Those tracking the money can’t outrun automation.

🔹 Claude agent uncovers Linux vulnerability; Red Hat has patched it, Ubuntu hasn’t
→ AI finds vulnerabilities faster than people can fix them.

🔹 OpenAI’s GPT-6.2 Astra reportedly enters limited testing
→ Just as its rival files a prospectus, OpenAI’s new model makes an appearance.

If AI really can improve itself, who would you want to hit the brakes?

Every day, I unpack one story worth your attention. Follow Dongye, and see you tomorrow.

#AI #Web3 #币安每小时充值量已超过提现量

Why are everyone’s points increasing so quickly? All of a sudden, everyone near the top of the rankings has tens of thousands of points. #币安神算子

FBI accuses North Korean hackers of stealing $6 billion from Binance; exchange resumes withdrawals

Binance suffered a 'TraderTraitor' hacking attack, with $6 billion stolen. The exchange processed more than $4.6 billion in withdrawals, while deposits reached $1.01 billion. Technical analysis should focus on market movements after the exchange resumed withdrawals. #Binance #ExchangeSecurity

Franklin Templeton Reveals: The Wealth Management Revolution of Next-Generation Financial Agents

Franklin Templeton’s latest report says that next-generation financial agents will lead a new trend in personal finance. As on-chain agent finance takes shape, how do these agents differ from general-purpose Personal Agents? Join the technical analysts for an in-depth look.

The FBI says North Korean hackers stole over $6 billion from Binance. The exchange has resumed withdrawals, but roughly half of the stolen funds have yet to be recovered.

The FBI attributed the attack to a North Korean group called TraderTraitor. Since resuming operations, Binance has processed $4.65 billion in withdrawals and $1.01 billion in deposits. Elliptic says about $3.24 billion remains in linked addresses, while $2.85 billion has been frozen or seized. Some funds have already entered mixers.

Why it matters
This is no longer just about one exchange. A state-sponsored hacking group has shown it can drain the hot wallet of the world’s largest exchange and start laundering the money within days. Resuming withdrawals was the easy part. Recovering the funds is a race against the mixers.
Custody is just a promise—and this week, it was put to the test.

4 more stories worth a look:
🔹 Hong Kong tests round-the-clock settlement and stablecoins
→ Banks that close at 5 p.m. are starting to look like a bug.
🔹 Perpetual DEX security: Can users exit on their own?
→ If withdrawals still require someone’s approval, the money isn’t really in your hands.
🔹 MIT Technology Review: The agent internet is here
→ Soon, the next visitor to your website might not be human.
🔹 Ethereum projects are moving back to L1
→ Taking the detour through rollups is getting more expensive.

One story, explained in depth every day. Follow Dongye, and see you tomorrow.

#AI #Web3 #币安已处理46.5亿美元提款