Frontier labs split on AI liability | Maine bans data centers

Bottom Line Up Front

Anthropic and OpenAI publicly split on an Illinois AI liability bill that would shield frontier developers from catastrophic-harm lawsuits.
Maine passes the first state-level ban on new data center construction, setting a potential model for other states.
The NAACP sues xAI in federal court over alleged Clean Air Act violations at its Memphis-area data center complex.
OpenAI launches GPT-5.4-Cyber, a restricted model for vetted security professionals with lowered refusal boundaries for defensive tasks.
Apple privately threatened to pull Grok from its App Store over nonconsensual deepfake nudes generated by the model.

I.AI Policy Today

Anthropic publicly opposes the Illinois AI liability bill that OpenAI supports

Anthropic told WIRED it opposes Illinois HB 5410, which would exempt frontier AI developers from civil liability for catastrophic harms if companies publish safety reports and lack intent to harm. The bill defines catastrophic harm as 100 or more deaths or serious injuries, or $1 billion or more in property damage. OpenAI testified in favor of the bill before the Illinois legislature on April 10, arguing a uniform standard would prevent a patchwork of state liability rules. HB 5410 is advancing through the Illinois General Assembly.

Read at WIRED ↗

Maine becomes the first US state to pass a data center construction ban

The Maine legislature voted to ban new data center construction, the first such state-level restriction in the US, the Financial Times reported. The measure comes three days after Sen. Sanders and Rep. Ocasio-Cortez introduced a federal AI Data Center Moratorium Act proposing a nationwide construction pause pending environmental and labor reviews. The ban may serve as a blueprint for other states considering moratoriums.

Read at Financial Times ↗

NAACP files Clean Air Act lawsuit against xAI over Memphis-area data center pollution

The NAACP filed suit in Mississippi federal court alleging that xAI's Colossus Gas Plant in Southaven, Mississippi, operated dozens of methane gas generators without required Clean Air Act permits to power Elon Musk's Memphis-area data center complex. The Southern Environmental Law Center and Earthjustice are representing the NAACP in the case. AIPD covered environmental groups' challenge to xAI's Mississippi data center permit on April 12; the NAACP's filing moves the dispute into federal court under the Clean Air Act.

Read at CNBC ↗

OpenAI launches GPT-5.4-Cyber, a restricted-access model for vetted security professionals

OpenAI released GPT-5.4-Cyber, a fine-tuned variant designed for defensive cybersecurity work with lower refusal boundaries for security tasks, including binary reverse engineering capabilities. The model is available only through OpenAI's Trusted Access for Cyber program, which has expanded to thousands of verified security professionals. The launch follows Anthropic's decision to restrict its Mythos model after testing revealed it could generate working exploits for zero-day vulnerabilities 72.4% of the time.

Read at SiliconANGLE ↗

Apple privately threatened to remove Grok from its App Store over deepfake nudes

Apple warned xAI behind closed doors that it would pull the Grok app from its App Store over nonconsensual sexual deepfakes generated by the AI model, NBC News reported. Apple confirmed the action in a letter to US senators, stating it demanded that developers "create a plan to improve content moderation." NBC News obtained the letter, which disclosed that the threat was made privately while the deepfake crisis on X played out publicly for weeks.

Read at NBC News ↗

Anthropic co-founder Jack Clark confirms the company briefed the Trump administration on Mythos

Anthropic co-founder and Head of Public Benefit Jack Clark confirmed at the Semafor World Economy summit that Anthropic voluntarily briefed the Trump administration on the Mythos model's cybersecurity capabilities. Anthropic is simultaneously suing the Department of Defense over the Pentagon's designation of Anthropic as a supply-chain risk, a label applied after the company refused to grant unrestricted military access to its AI models.

Read at TechCrunch ↗

II.China Watch

Unitree Robotics begins global humanoid robot sales ahead of IPO

Hangzhou-based Unitree Robotics began selling its R1 humanoid robot on Alibaba's AliExpress platform, pricing units at roughly $4,400 to $8,200 with free US shipping. The company targets North America, Europe, Japan, and Singapore, per Bloomberg. Unitree shipped over 5,500 humanoid robots in 2025, far outpacing Tesla, Figure AI, and Agility Robotics at roughly 150 units each, and aims for 10,000 to 20,000 units in 2026. Over 80% of the R1's components are sourced from Chinese suppliers, and the direct-to-consumer export model brings dual-use robotic hardware into consumer channels without existing export control classification.

Read at Bloomberg ↗

Iran used Chinese-built spy satellite to target US military bases during March strikes

Leaked Iranian military documents obtained by the Financial Times show the Islamic Revolutionary Guard Corps (IRGC) Aerospace Force secretly acquired a surveillance satellite, the TEE-01B, built and launched by Chinese company Earth Eye Co, and used it to capture imagery of US military installations across the Middle East. The satellite photographed Prince Sultan Air Base in Saudi Arabia, which was struck shortly after satellite passes. Chinese firm Emposat provided the IRGC with ground-station software and a global network to direct the satellite's operations.

Read at Financial Times ↗

Former UN AI adviser tells UK MPs China is now the "good guy" on AI governance

Prof Dame Wendy Hall, a former member of the UN's AI advisory board who co-wrote a UK government AI review, told the House of Commons business and trade committee that China is backing multilateral AI governance attempts while the US pursues what she called a "wild west" approach, per the Guardian. Hall testified that China's regulatory framework for AI is more developed than Washington's deregulatory stance under the Trump administration. Hall characterized the US approach as driven by "profit-hungry companies" operating without adequate regulatory guardrails.

Read at The Guardian ↗

III.Federal Policy Tracker

State Department exploring agentic AI for high-volume diplomatic workflows

Amy Ritualo, the State Department's acting chief data and AI officer, described agentic AI as "the executive assistant" for handling high-volume workflows at an AITalks event, per FedScoop. Ritualo said the department is at an "exploration stage" for agentic AI and is focused on establishing guardrails and incorporating workforce input before full implementation. The State Department has not yet set a deployment timeline for autonomous AI agents in its operations.

Read at FedScoop ↗

Democrats warn Big Tech lobby groups are deploying millions ahead of AI regulation fights

Congressional Democrats are raising alarms about the scale of industry lobbying on AI policy, with pro-industry campaign groups deploying millions amid growing public support for tighter AI regulation, the Financial Times reported. The lobbying push comes as frontier AI companies take divergent positions on state liability legislation, with OpenAI and Anthropic publicly split on Illinois HB 5410. The FT reported that lobby groups are deploying funds specifically to shape the terms of federal AI regulation before Congress acts.

Read at Financial Times ↗

IV.Capability & Research Watch

Researchers hijack AI coding agents from Anthropic, Google, and Microsoft to steal developer credentials

Security researchers demonstrated that prompt injection attacks can hijack Claude Code Security Review, Gemini CLI Action, and GitHub Copilot through their GitHub Actions integrations, stealing API keys and access tokens from developers, per The Register. All three vendors paid bug bounties to the researchers but none assigned CVEs or published public security advisories disclosing the vulnerabilities. The researchers warned the problem is "probably pervasive" across AI agents with code repository access.

Read at The Register ↗

Deepfake nudes have hit nearly 90 schools and 600 students worldwide

A WIRED and Indicator analysis identified nearly 90 schools and approximately 600 students globally targeted by AI-generated deepfake nude images. Multiple US state legislatures are considering or have passed bills criminalizing nonconsensual AI-generated intimate imagery. The count is the first systematic tally of school-level deepfake incidents, a data gap that legislators have cited when debating the scope of the problem.

Read at WIRED ↗

V.Industry & Market Watch

Snap cuts up to 16% of its workforce, citing AI-driven efficiencies

Snap announced plans to cut up to 16% of its global workforce, attributing the layoffs to AI-enabled productivity gains that allow the company to accomplish more with fewer employees, CNBC reported. The announcement is among the first by a major tech company to directly name AI as the reason for large-scale layoffs, providing a concrete data point for congressional debates over AI workforce displacement.

Read at CNBC ↗

SoftBank expands syndication for $40 billion loan backing its OpenAI investment

SoftBank's lenders are inviting additional banks to join a $40 billion loan that finances SoftBank's investment in OpenAI, Bloomberg reported. The $40 billion facility is among the largest single debt instruments backing an AI company. The loan expansion tests creditor appetite for concentrated AI exposure. Moody's Analytics separately flagged "emerging signs of broader strain among exposed lenders" in a report published the same day.

Read at Bloomberg ↗

Uber commits $10 billion to robotaxis in strategy reversal

Uber announced a $10 billion commitment to robotaxis through equity investments and vehicle orders, reversing its previous decision to exit autonomous vehicles after selling its self-driving unit, the Financial Times reported. The company described the strategy as an effort to make up lost ground against Waymo and other autonomous vehicle operators. Uber's return to autonomous vehicles will require regulatory approvals across its US markets, where it currently operates in more than 10,000 cities.

Read at Financial Times ↗