AI Policy · Daily

The FTC began enforcing the TAKE IT DOWN Act against 15 platforms Tuesday, opening a complaint portal and a 48-hour takedown clock for nonconsensual intimate imagery. A D.C. Circuit panel split Tuesday while weighing Anthropic's challenge to Defense Secretary Pete Hegseth's supply chain risk designation, with one judge calling it overreach and two Trump appointees probing the deference owed to the Pentagon. The Pentagon awarded Shield AI a contract to integrate its Hivemind autonomy software onto the Low-Cost Uncrewed Combat Attack System (LUCAS) for coordinated drone swarm operations. The USDA Inspector General reported the department rolled out AI for supply chain, crop yield and permitting work without required cybersecurity controls or any generative AI policy.

I.AI Policy Today

FTC opens TAKE IT DOWN Act enforcement against 15 platforms with 48-hour takedown clock

The Federal Trade Commission began enforcing the TAKE IT DOWN Act on Tuesday and launched a public complaint portal at takeitdown.ftc.gov, the agency said in a press release. The law requires covered platforms to remove nonconsensual intimate imagery, including AI generated deepfakes, within 48 hours of receiving a verified victim notice. FTC Chairman Andrew Ferguson previously sent letters to 15 platforms including Amazon, Alphabet, Apple, Meta, Microsoft, Reddit, Snapchat and TikTok setting the May 19 compliance deadline, as reported by the FTC in AIPD's May 12th edition.

Read at FTC ↗

D.C. Circuit panel divided over Anthropic challenge to Pentagon supply chain risk designation

A three-judge panel of the U.S. Court of Appeals for the District of Columbia Circuit heard oral argument Tuesday in Anthropic's challenge to Defense Secretary Pete Hegseth's designation of the company as a supply chain risk, AP reported. Judge Karen LeCraft Henderson, a George H.W. Bush nominee, called the designation "spectacular overreach by the [Defense] Department." Judges Neomi Rao and Gregory Katsas, both Trump nominees, pressed Anthropic counsel Kelly Dunbar on the deference owed to the secretary's risk judgment. Justice Department attorney Sharon Swingle told the court the failure of an AI model in active military operations could carry catastrophic national security consequences. Dunbar argued the designation "defied congressionally mandated procedures, exceeded statutory limits and violated the Constitution," the Washington Post reported. Earlier this month the D.C. Circuit rejected Anthropic's request for an order blocking the Pentagon's actions during the appeal, and a federal judge in San Francisco ruled in Anthropic's favor in March in a parallel case.

Read at The Hill ↗ Read at AP ↗ Read at Washington Post ↗

Pentagon picks Shield AI's Hivemind software to fly LUCAS attack drone swarms

The Department of Defense awarded Shield AI a contract to integrate its Hivemind autonomy software onto the new Low-Cost Uncrewed Combat Attack System (LUCAS) as part of a pilot program for drone swarms, Breaking Defense reported. The $35,000, 10-foot LUCAS is reverse-engineered from the Iranian Shahed 136 and carries an explosive payload that detonates on impact. Once Hivemind is integrated, groups of LUCAS drones are to coordinate and maneuver without human intervention while a single operator retains strike authority. Shield AI said it expects to participate in an operational swarm demonstration this fall. The company recently closed a $2 billion funding round at a $12.7 billion valuation, CNBC reported. The Defense Department has detailed plans to spend nearly $55 billion on drone and autonomy development in fiscal 2027 under the Defense Autonomous Warfare Group, which absorbed the Biden era Replicator initiative.

Read at CNBC ↗ Read at Breaking Defense ↗

USDA Inspector General finds department deployed AI without required cybersecurity or governance controls

The USDA Office of Inspector General released a report finding the department lacks required cybersecurity and governance controls on its AI systems, GovExec reported. The IG found USDA has no generative AI policy. USDA uses AI for supply chain risk detection, crop yield estimation and permitting recommendations. The IG attributed the gap to the department prioritizing AI rollout over controls.

Read at GovExec ↗

Senate Republicans put $108M for 200 DHS CSAM specialists into the reconciliation bill

Senate Republicans on Tuesday added a $108 million provision over three years to the reconciliation bill funding Immigration and Customs Enforcement and Border Patrol, the New York Times reported. The provision would expand specialists combating child sexual abuse material and trafficking at the Department of Homeland Security's investigations division to 200 from seven. The measure is championed by Sen. Josh Hawley (R-Mo.) and is intended to scale forensic and victim identification work targeting AI generated deepfake and other CSAM. The reconciliation bill is expected to pass the Senate as soon as this week. The House is expected to take up the Senate measure by June 1, a deadline set by President Trump, GV Wire reported. Hawley said the funding aims to help identify an estimated 89,000 children seen in online sexual abuse material who remain unidentified.

Read at NYT ↗ Read at GV Wire ↗

II.China Watch

Beijing confirms U.S.-China AI governance dialogue track

Foreign Ministry spokesman Guo Jiakun on Tuesday said the two governments will hold further dialogues on AI governance following exchanges between Presidents Xi Jinping and Donald Trump during last week's state visit, the South China Morning Post reported. Guo said the two leaders had agreed their governments should keep talking on managing the risks of advanced AI systems. Treasury Secretary Scott Bessent had first signaled the U.S.-China AI safety talks before the Beijing summit, as reported by the New York Times in AIPD's May 15th edition.

Read at SCMP China ↗

Alibaba releases Qwen3.7-Max flagship model for agentic workflows

Alibaba's Qwen team on Wednesday released Qwen3.7-Max, a new flagship large language model the company said offers stronger long-cycle autonomous execution for agentic workflows, PingWest reported. The model will be available through Alibaba Cloud's Bailian enterprise API in China. The launch came during Alibaba's annual Cloud Summit in Hangzhou.

Read at PingWest ↗

AMD CEO Lisa Su praises China AI sector at Shanghai DevDay

AMD CEO Lisa Su praised the openness of China's AI sector at the company's AI Developer Day in Shanghai on May 19, Caixin reported. Su said AMD will provide full-stack compute support for local compute and edge inference through multi-agent architectures. She predicted global active AI users will reach 5 billion in the coming years. She added that GPUs will be deployed beyond the cloud, and that the spread of edge agent deployments will create new requirements on CPU compute.

Read at Caixin ↗

YMTC begins pre-IPO coaching for STAR Market listing

Yangtze Memory Technologies Holding Co., China's leading NAND flash memory maker, has completed IPO tutoring registration with the Hubei branch of the China Securities Regulatory Commission, with CITIC Securities serving as sponsor, TechNode reported. The filing positions YMTC for a Shanghai STAR Market debut by the country's only mainland-based integrated device manufacturer with full 3D NAND production capabilities. YMTC has sat on the Bureau of Industry and Security's Entity List since December 2022, restricting its access to advanced U.S. semiconductor manufacturing equipment.

Read at TechNode ↗

III.Capability & Research Watch

OpenAI adds C2PA conformance and Google SynthID watermarking to ChatGPT, Codex and API images

OpenAI rolled out a multilayered content provenance approach combining C2PA conformance, SynthID watermarking via a partnership with Google DeepMind, and a public verification tool for AI generated images. OpenAI joined the Steering Committee of the Coalition for Content Provenance and Authenticity in 2024 when it began adding Content Credentials to DALL-E 3 outputs. It has now become a C2PA Conforming Generator Product. SynthID embeds an invisible watermark in images generated through ChatGPT, Codex or the OpenAI API to complement C2PA metadata that can be stripped by uploads, format changes or screenshots. The public verification tool is limited at launch to OpenAI generated content. The company said it aims to support cross-platform verification "in the upcoming months."

Read at OpenAI ↗

Independent study reports GPT-5 and Claude lines score near zero on agentic harm scenarios while DeepSeek V3.2 fails catastrophically

An independent agentic misalignment study tested 22 models from nine developers across blackmail, espionage and murder scenarios under five instruction conditions, the researchers wrote on LessWrong. GPT-5.4, GPT-5.5, Claude Sonnet 4.6, Opus 4.6 and Opus 4.7 scored between 0% and 1% across all conditions. DeepSeek V3.2 took the murder action at 100%, leaked at 98% and blackmailed at 94%. The work was funded by a BlueDot Impact rapid grant and published as the sequel to "Blackmail at 8 Billion Parameters."

Read at LessWrong ↗

IV.Industry & Market Watch

Meta lays off 8,000 employees and shifts 7,000 staff into four AI units

Meta began notifying around 8,000 employees of layoffs on Wednesday and is shifting roughly 7,000 staff into four new AI units, the New York Times reported. Chief People Officer Janelle Gale told staff in an internal memo that the cuts are part of "continued effort to run the company more efficiently and to allow us to offset the other investments we're making." Affected U.S. employees will receive 16 weeks of base pay plus tenure-based severance and extended healthcare coverage. The company had nearly 79,000 employees at the start of the year and expects capital expenditures to rise by at least 60% this year over 2025 to support its Meta Superintelligence Labs efforts, Axios reported.

Read at NYT ↗ Read at Axios ↗

OpenAI commits $234M to first applied AI lab outside the U.S., partnering with Singapore government

OpenAI committed more than S$300 million ($234 million) over multiple years to open its first applied AI lab outside the United States in Singapore through a memorandum of understanding with the Ministry of Digital Development and Information, Bloomberg reported. The company plans to grow its Singapore-based technical team to more than 200 roles over the next few years, expanding an office it first established in 2024, Channel News Asia reported. The MOU also covers a Singapore chapter of the OpenAI Academy, Codex for Teachers hackathons and collaboration with the Infocomm Media Development Authority (IMDA) and AI Singapore on the AIxTech program. Singapore signed a parallel AI partnership with Google at the same ATX Summit announcement.

Read at Bloomberg ↗ Read at CNA ↗

Samsung union confirms 48,000-worker chip strike for Thursday after wage talks collapse

Samsung Electronics' union plans for 48,000 workers, mostly in the company's chip divisions, to walk off the job Thursday after the latest round of wage talks fell through on bonus payments, AP reported. Union leader Choi Seung-ho said unionized workers will begin an 18-day strike from Thursday, per WKZO. South Korean Prime Minister Kim Min-seok said the strike could cause up to 100 trillion won ($66 billion) in economic damage by disrupting Samsung's semiconductor manufacturing. Samsung said operating profit for the January-March quarter jumped eightfold to a record 57.2 trillion won ($38 billion). The union has said compensation has lagged despite the company's AI driven memory chip surge. South Korea's central bank estimated a worst-case strike could shave 0.5 percentage points off a forecast 2.0% expansion for the South Korean economy this year.

Read at AP ↗ Read at WKZO ↗

Ex-OpenAI staffers launch Guidelight AI Standards, urge SpaceX investors to demand xAI safety disclosures ahead of IPO

Guidelight AI Standards, a new AI safety nonprofit cofounded by former OpenAI safety researcher Steven Adler and former OpenAI policy adviser Page Hedley, joined Legal Advocates for Safe Science and Technology, Encode AI and The Midas Project on a Tuesday open letter to SpaceX investors warning of "unpriced risks" tied to xAI, Wired reported. SpaceX is preparing an IPO that could raise up to $75 billion at a private valuation over $1 trillion, after acquiring xAI last year. The signatories cite specific xAI incidents including Grok producing white genocide content and sexualized images of women and children that prompted 37 U.S. attorneys general to demand corrective steps. The letter calls on SpaceX to disclose whether xAI will continue developing frontier AI models and to publish a public safety and governance plan; the Washington Post previously reported xAI had only "two or three" people working on safety as of January.

Read at Wired ↗

Dell unveils Deskside Agentic AI as it pitches on-premise compute against rising cloud token costs

Dell COO Jeff Clarke said token consumption for AI reasoning is up 320x as agentic systems break cloud economics, Forbes reported in coverage of Dell Technologies World 2026. Dell announced Dell Deskside Agentic AI, an on-premise sandbox for building and running AI agents locally, Computer Weekly reported. The product is powered by Nvidia NemoClaw on Dell workstations supporting models from 30 billion up to a trillion parameters. Dell senior vice president Jon Siegal said a single Dell developer recently burned through 1 billion tokens in 24 hours, racking up a $3,400 cloud bill in one day. Dell also said Google Gemini models will run on-premise via Google Distributed Cloud on Dell PowerEdge servers. Parallel partnerships will bring Palantir Foundry and SpaceX AI's Grok on-premise.

Read at Forbes ↗ Read at Computer Weekly ↗

V.Global & Geopolitics

NSPCC-led coalition urges PM Starmer to limit under-16 social media ban to apps with "risky" features

The NSPCC, Molly Rose Foundation and Smartphone Free Childhood wrote to Prime Minister Keir Starmer on Wednesday urging Britain to require strict safety standards for tech platforms serving under-16s rather than impose a broader Australia-style ban, The Guardian reported. The coalition called for restrictions on platform features such as infinite scrolling, disappearing messages and push notifications. The letter framed a formal age limit as a "cliff edge" safety risk. The UK government consultation on new online safety measures closes in a week. Communications watchdog Ofcom oversees the Online Safety Act.

Read at The Guardian ↗

Former Google DeepMind AI engineer files UK tribunal unfair dismissal claim over Israel work

A former Google DeepMind AI engineer filed an unfair dismissal tribunal claim in the UK, saying he was dismissed after protesting the company's work for the Israeli government, The Guardian reported. The engineer distributed flyers in DeepMind's London office reading "Google provides military AI to forces committing genocide" and emailed colleagues about Google's 2025 decision to drop its prior pledge not to pursue weapons that harm people. Google has disputed the allegations.

Read at The Guardian ↗