1｜Anthropic Calls for Leading AI Labs to Establish Pause Mechanism

On June 5, Anthropic called for major AI labs to implement a pause mechanism for advanced development, warning that humans could lose control of AI systems. The specific risk highlighted is recursive self-improvement, where models begin to improve themselves in ways that exceed human oversight. Both Reuters and Al Jazeera positioned this news at the core of tech regulation, as it is not just a standard safety commitment but a top lab’s request for peers to accept brake rules.

What truly shakes up the industry power structure is this signal. Over the past two years, AI companies have leveraged safety more as a bargaining chip in policy discussions. Now, Anthropic is taking the issue a step further: as models approach self-iterative capabilities, who can command labs to halt. Governments, boards of directors, competitors, users—all lack a clear answer. The pause mechanism appears to be a safety tool on the surface, but at its core, it’s about the allocation of control in advanced AI.

2｜US Debates State AI Regulations While Discussing Investment in AI Companies

Members of the US House of Representatives have introduced a bill attempting to ban state-level AI regulations. This move appears to be about unified regulation, but underneath, it’s a struggle for federal government dominance over AI rules. New York and Seattle are using data center moratoriums to demonstrate that states and cities also want to engage in AI governance. The congressional bill is equivalent to first pinning down local regulations.

More critically, there’s a capital inflow aspect. According to The Information and WSJ headlines, US officials are discussing investing in AI companies. Regulation and equity are part of the same set of signals, indicating that AI is transitioning from “an industry needing regulation” to “infrastructure needing national investment.” This deviates from traditional tech regulation logic and leans more towards semiconductors, energy, and defense industries. The government not only seeks to define boundaries but also aims to secure future returns and strategic control.

3｜New York and Seattle Halt Data Centers, While Meta Expands Using Tents

The New York State Assembly has passed a one-year moratorium on large data centers, which would become the first state-level ban of its kind in the U.S. Seattle is also preparing to enact a one-year pause on AI data centers to study their community impact. Local governments are concerned about power, water, land, and pollution. They see not an AI revolution, but bills and infrastructure strain.

On the other hand, Meta has been reported to be setting up temporary server tents in multiple locations in the U.S., which can be built in three months and powered by jet engines. On the same day, AirTrunk announced a $30 billion investment in India to build 5GW of AI data center capacity. Two movements in AI infrastructure are happening simultaneously. Capital wants to push the pace of AI construction to the limit, while local politics are starting to demand a pause to calculate the external costs first. Data centers are no longer just tech assets; they are becoming issues of city governance and energy distribution.

4｜South Korea Demands Sharing of AI Windfall Profits, Japan Worries about Becoming an AI Colony

The South Korean Minister of Labor has called on tech companies to share the windfall profits from AI with suppliers and employees. This statement shifts AI from being an enterprise efficiency tool to a labor distribution issue. Past policy discussions have mostly revolved around unemployment benefits and skills training, but this time South Korea is asking a more direct question: if AI allows platform companies to earn excess profits, should the supply chain and employees receive a portion of it.

The warning from the Japanese Minister of Digital indicates a more dire situation. He said if Japan falls behind, it could become an “AI colony.” CNBC mentioned on the same day that China is attracting more American AI talent back, aiming for the next generation of super applications. These three threads together illustrate that AI sovereignty is no longer just about domesticating models. It also includes talent flow, profit distribution, and application entry points. Whoever defines the foundational model is closer to the upstream of the next industrial order.

5｜Three Major Banks to Co-Build Tokenized Deposit Network, Stablecoin Competition Enters Banking Internally

JPMorgan Chase, Bank of America, and Citi are preparing to co-build a shared tokenized deposit network, planning to launch in early 2027. CoinDesk states that this is not banks catching up to crypto-native projects, but banks using their deposit liabilities to redesign on-chain settlements. Stablecoins have been viewed in the past as payment tools outside the banking system; now, large banks want to stuff the same real-time settlement capabilities back into regulated deposit networks.

This state-level regulatory tug-of-war is connected to the FedAccounts and Stablecoin Legislation. Payments Dive reports that several U.S. states still want to retain regulatory authority over stablecoins. Traditional financial players have a clear strategy. They may not necessarily want to block stablecoins, but rather to unpack the core capabilities of stablecoins and fit them back into the banking and regulatory framework. In the next round of competition in DeFi, it may not be USDC against bank deposits, but rather bank deposits themselves becoming on-chain assets.

Also worth noting:

Anthropic eases tensions with the White House, cooling regulatory relations ahead of IPO. The company, previously on a federal watchlist, is realigning with the government. As AI companies emphasize security more, they need to demonstrate they can still fit into the national security apparatus.

NSA reportedly preparing to deploy Anthropic Mythos for cyber operations. Federal bans and national security needs clash, indicating the shift of the most advanced models from commercial tools to strategic assets.

AI data center memory consumption starting to squeeze other industries. Industry alliances urge the Trump administration to act, warning that AI-driven memory shortages could hike costs in automotive, healthcare, and telecom sectors. The computational power inflation is spilling over into non-AI industries.

Switch in talks for funding at a valuation of over $50 billion. Data center developers’ valuations continue to be driven up by AI infrastructure, with companies that sell shovels becoming the most stable assets in this cycle.

Airbnb CEO Brian Chesky plans to establish a new AI lab. Airbnb declined direct LLM partnerships last year, and now aims to bring AI capabilities back in-house. App companies are beginning to realize that the user experience at the entry point cannot be entirely handed over to external model providers.

Quantum Bit announces the world’s first robot training complex opens, with 300,000 Chinese residences available for robot training. Embodied intelligence moves from the lab to real space, with training grounds themselves becoming a form of infrastructure asset.

[BlockBeats]