AI Policy · Daily

  • Anthropic's Project Glasswing unites 12 companies for defensive cybersecurity while the Pentagon maintains its Anthropic blacklisting, a policy contradiction within the same administration.
  • San Francisco police arrest two suspects in a second attack on OpenAI CEO Sam Altman's Russian Hill home in three days, following a Molotov cocktail attack on April 10.
  • AI executives including Sam Altman and Dario Amodei are rolling out sweeping regulatory proposals, but congressional gridlock may stall their Washington ambitions — with OpenAI already pivoting to state-level advocacy.

I.AI Policy Today

Trump administration pushes banks to test Anthropic's Mythos while Pentagon maintains blacklisting

The Trump administration is encouraging major banks to test Anthropic's Mythos model for cybersecurity while the Pentagon maintains the company's supply-chain risk designation, per TechCrunch citing Bloomberg. JPMorgan Chase, Goldman Sachs, Citigroup, Bank of America, and Morgan Stanley are reportedly testing Mythos for vulnerability detection at the urging of Treasury Secretary Bessent and Fed Chair Powell. A federal court denied Anthropic's request for a stay of the Pentagon designation but agreed to expedite the hearing, per Inside AI Policy.

Read at TechCrunch ↗

AI CEOs' Washington policy ambitions may stall against Congress's track record

AI executives including OpenAI's Sam Altman and Anthropic's Dario Amodei are rolling out sweeping policy proposals to shape how their products are regulated, but Axios reports the pitches may face the same gridlock that has stalled tech legislation for years. OpenAI's chief global affairs officer Chris Lehane told Axios the company is focusing policy efforts at the state level, where electoral dynamics give advocates more leverage. Anthropic, meanwhile, is expanding its D.C. presence under new public policy head Sarah Heck, a former Stripe and Obama NSC official.

Read at Axios ↗

AI companies fund think tanks and policy papers to counter public image problems

OpenAI and Anthropic are aggressively backing policy initiatives and think tanks as public trust in AI declines, per The Guardian. OpenAI released a 13-page industrial policy paper advocating for a four-day workweek and a public wealth fund to distribute AI profits, while Anthropic announced the creation of the Anthropic Institute to examine AI's societal disruptions. Critics including the AI Now Institute's Sarah Myers West argue the proposals frame AI as an inevitable force rather than a regulatable product, effectively sidestepping corporate accountability.

Read at The Guardian ↗

II.China Watch

MiniMax open-sources M2.7 model with day-one domestic chip support

Chinese AI startup MiniMax open-sourced its large language model M2.7 on April 12, per PingWest. The model rivals leading US models on software engineering benchmarks, scoring 56.22% on SWE-Pro and described as a self-evolving agent capable of autonomous task completion. MiniMax launched M2.7 with first-day adaptation for Huawei Ascend, Moore Threads, Metax, and Kunlun Chip alongside NVIDIA GPUs.

Read at PingWest ↗

Tsinghua AI researcher Wu Yi departs for Meta's Superintelligence Labs

Wu Yi, an assistant professor at Tsinghua University's Institute for Interdisciplinary Information Sciences, has left his faculty position to join Meta's Superintelligence Labs, per QbitAI. Meta CEO Mark Zuckerberg personally recruited Wu Yi to work on superintelligence and multi-agent systems at MSL, according to the report. Tsinghua IIIS has already removed Wu Yi's information from its official website.

Read at QbitAI ↗

Agibot releases zero-code robotics deployment platform

Chinese embodied AI firm Agibot launched Genie Studio Agent, a zero-code application platform covering vision-language-action (VLA) models, reinforcement learning, and visual perception, per PingWest. Users can orchestrate robot workflows through a drag-and-drop interface without writing code. Agibot's GO-2 foundation model leads the LIBERO robotics benchmark with a 98.5% average success rate.

Read at PingWest ↗

III.Capability & Research Watch

Linux 7.0 launches with official Rust support as Torvalds weighs AI's impact on kernel security

Linux kernel 7.0 released today with official Rust language support, making Rust a permanent part of kernel development, per The Register. Creator Linus Torvalds publicly questioned whether AI-driven bug-finding tools should change the kernel release process, citing the recent wave of AI-discovered vulnerabilities. Rust support has been in testing since Linux 6.1 and is now fully integrated into the mainline kernel.

Read at The Register ↗

Scientists question whether Anthropic's Mythos claims are safety disclosure or marketing

Scientists are criticizing Anthropic for using "purposely vague language that obscures evidence" about Mythos's capabilities, The Guardian reported. Skeptics suggest the company's decision to withhold the model from public release may be designed to attract investment without independent scrutiny. Independent cybersecurity researchers have not been able to verify Anthropic's specific vulnerability claims, according to the report.

Read at The Guardian ↗

IV.Global & Geopolitics

UK financial regulators rush to assess Mythos cybersecurity risks for banks, insurers, and exchanges

UK financial regulators are assessing cybersecurity risks posed by Anthropic's Claude Mythos model, the Financial Times reported. The Bank of England, Financial Conduct Authority, and Treasury officials are holding emergency talks with the National Cyber Security Centre, with major banks, insurers, and exchanges expected to be briefed at a dedicated meeting within a fortnight. The Bank of Canada held parallel discussions with its financial sector last week, coinciding with the US Treasury and Federal Reserve encouraging American banks to test Mythos for defensive cybersecurity.

Read at Financial Times ↗

OpenAI opens first permanent London office days after pausing UK Stargate

OpenAI announced its first permanent London office, days after halting the UK Stargate data center expansion over energy costs, per CNBC. OpenAI designated London as its largest non-US research hub, and the new 88,500 sq ft space at Regent Quarter is sized to more than double the company's current UK headcount. The office opening redirects OpenAI's UK investment from infrastructure to research talent.

Read at CNBC ↗