AI Policy · Daily

House Energy & Commerce Committee Ranking Member Frank Pallone called for a national moratorium on AI data centers before a House Energy subcommittee markup, citing grid strain and slow federal interconnection reforms. The White House has rerouted Anthropic negotiations over restoring Claude Fable 5 access to cofounder Tom Brown, sidelining CEO Dario Amodei. Anthropic told the Senate Banking Committee that Alibaba-linked operators ran 28.8 million exchanges through 25,000 fraudulent accounts to copy Claude's capabilities, two days before Commerce restricted exports of its Mythos and Fable models. Chinese AI researchers share U.S. "Chernobyl moment" anxieties.

I.Top Stories

Pallone, top House Energy Democrat, calls for a national AI data center moratorium at Wednesday markup

Rep. Frank Pallone (D-N.J.), the ranking Democrat on the House Energy and Commerce Committee, called for "a national AI data center moratorium" before a Wednesday afternoon energy subcommittee markup of eight electricity and pipeline safety bills, including the Ratepayer Protection Act. Pallone described the proposals being marked up as a useful first step but not nearly enough against the strain data centers are putting on the grid. He said Federal Energy Regulatory Commission interconnection reforms would take months or years to take effect, per Heatmap News. He cited a resolution in Asbury Park, N.J., calling for a statewide moratorium and the city of New Brunswick rejecting a planned data center project this year. Pallone is the highest-ranking Democratic energy policy voice to back a position previously held mainly by Sen. Bernie Sanders (I-Vt.) and Rep. Alexandria Ocasio-Cortez (D-N.Y.), whose Artificial Intelligence Data Center Moratorium Act would tie any restart of construction to passage of federal AI legislation.

Read at Heatmap News ↗ • Read at The Hill ↗ • Read at House Committee on Energy and Commerce ↗

Trump White House routes Anthropic talks through cofounder Tom Brown rather than CEO Amodei, Wired reports

The Trump administration has been negotiating Anthropic's re-release of Claude Fable 5 with cofounder Tom Brown rather than CEO Dario Amodei, with one person familiar with the calls telling Wired that Brown "can actually engage" while calling Amodei a "weirdo." Anthropic's outreach has been led by Brown and public policy chief Sarah Heck, with both leadership and working-group calls in recent days about what proof would alleviate administration concerns about jailbreaks of Fable 5. Reps. Sam Liccardo, Jay Obernolte, C. Scott Franklin and Ted Lieu sent Commerce Secretary Howard Lutnick a bipartisan letter the previous week asking for specific criteria and a timeline for restoring access.

Read at Wired ↗

Anthropic accuses Alibaba of "largest known distillation attack" on Claude in letter to Senate Banking

Anthropic told Senate Banking Committee Chair Tim Scott and Ranking Member Elizabeth Warren in a June 10 letter that operators affiliated with Alibaba and its Qwen AI lab ran "the largest known distillation attack" on Claude. The campaign generated more than 28.8 million exchanges through nearly 25,000 fraudulent accounts between April 22 and June 5, per a Reuters wire. Distillation trains a smaller model on a stronger model's outputs. Anthropic said the campaign was a route for China to accelerate toward its Mythos Preview capabilities and that it had previously identified similar efforts by DeepSeek, Moonshot AI and MiniMax in a February posting. The Commerce Department imposed export restrictions on Anthropic's Mythos and Fable models on June 12, two days after the letter was sent, and Alibaba shares slid to a 16-month low in Hong Kong after Bloomberg and the WSJ reported the accusations.

Read at Reuters ↗ • Read at Bloomberg ↗ • Read at WSJ ↗

Chinese AI researchers share U.S. "Chernobyl moment" anxieties at Beijing conference, Wired reports

At a Beijing Academy of Artificial Intelligence conference in Zhongguancun, Chinese AI researchers voiced concerns about catastrophic AI risks closely mirroring those of U.S. counterparts, with both sides describing a possible "Chernobyl moment" if frontier development goes wrong, Wired correspondent Will Knight reported. Sessions covered recursive self-improvement and humanoid robotics, with attendees including Whitfield Diffie, co-inventor of public-key cryptography, and Andrew Barto, who won the Turing Award with Rich Sutton for reinforcement learning. MIT computer scientist Stephen Casper, who spoke via video, told Wired that the benefits of international AI safety collaboration outweigh the national security risks. Knight wrote that the U.S. and China should put their AI rivalry aside given the cybersecurity and systemic risks of frontier AI.

Read at Wired ↗

II.China Watch

ByteDance launches flagship Doubao model and frames cost engineering as China's answer to U.S. chip controls

ByteDance's Volcano Engine unveiled the Doubao 2.1 Pro flagship at its FORCE Conference, with daily token calls across the Doubao family now reaching 180 trillion, per Pandaily. Volcano Engine President Tan Dai said the total cost of running Doubao 2.1 Pro is about 20% of Anthropic's Claude Opus 4.6, achieved through quantization, distillation and proprietary chip development. ByteDance also rolled out a paid Doubao Pro consumer subscription starting at 68 yuan (about $10) a month with a new office task mode, the first major monetization push for what QuestMobile ranks as China's most used consumer AI app. Tan Dai attributed the cost advantage to in-house silicon, a structural response that has become central to Chinese model vendors' positioning as Commerce Department export controls continue to restrict their access to top tier U.S. AI accelerators.

Read at Pandaily ↗

Tencent launches an email service built for AI agents with their own digital identity

Tencent has launched Agently Mail, a dedicated email service that gives AI agents their own digital identities isolated from users' personal QQ Mail accounts, per Pandaily. Agents can register independently, receive verification codes, and conduct business-to-business workflows including inquiries, quotations, order placement and invoice reconciliation on their own behalf. Registration requires real-name authentication and the platform supports integration with ByteDance's Doubao, Moonshot AI's Kimi Work, Anthropic's Claude Code and other domestic and foreign agent systems. The isolation, real-name and audit-logging design extends the Cyberspace Administration of China's online traceability requirements down to the agent identity layer, an industry-led answer to a governance question that no current Chinese rule directly addresses.

Read at Pandaily ↗

Self-driving leader Momenta clears Hong Kong IPO hearing and rebrands as a "physical AI" company

Momenta, the assisted-driving software company that supplies systems to nine of the world's 10 largest carmakers, cleared its Hong Kong Stock Exchange listing hearing on June 23 and is targeting roughly $1 billion in proceeds at a valuation near $14.8 billion, per QbitAI. The company has repositioned from autonomous driving to "physical AI," a category built on world models that generate predictions and plans for action in physical environments. Momenta's R7 world model entered mass production in April on SAIC-Volkswagen's ID. ERA 9X, drawing on driving data from a fleet that has logged more than 12 billion kilometers. A successful listing would make Momenta the first Hong Kong-listed pure-play physical AI stock and channel mainland capital into the world model and embodied AI infrastructure that Beijing has flagged as strategic for the next phase of AI competition.

Read at QbitAI ↗

III.Policy Tracker

House Republican denies AI authorship of NDAA amendment after Claude screenshots circulate

Rep. Anna Paulina Luna (R-Fla.) said her staff used AI only for "spellcheck" on the summary of an amendment to the 2027 National Defense Authorization Act, denying AI was used for the bill text, after X users circulated screenshots of an amendment summary containing the phrase "Claude responded:" followed by amendment summary text, per The Verge. Luna initially said staff used AI to correct draft text without editing, then edited her response to specify that AI touched only the amendment summary, not the bill text. She said all House bill text comes from the Office of the Legislative Counsel, which she said is prohibited from using AI. The amendment was described as identical to a bill from the 118th Congress requiring the Defense Secretary to designate Department of Defense activities at the southwest land border as a named operation.

Read at The Verge ↗

ICE and CBP AI surveillance contracts hit $513 million in 2026, new report finds

A new report from immigration rights group Mijente, Just Futures Law and Surveillance Resistance Lab finds that U.S. Immigration and Customs Enforcement and Customs and Border Protection contracts with 11 named AI surveillance vendors reached a record $513 million in 2026, up from about $310 million in 2025 and under $50 million in 2013, per The Guardian. The growth is driven largely by new contracts for Palantir, the data analytics firm central to ICE enforcement, and Anduril, the defense company that builds AI surveillance towers, drones and sensors. The portfolio covers data brokers, social media scrapers, facial recognition, phone hacking devices, spyware, and external contractors the authors characterize as "bounty hunters."

Read at The Guardian ↗

RAISE US, a bipartisan state-employer AI workforce coalition led by Raimondo and Holcomb, launches Thursday

A bipartisan consortium called RAISE US launches Thursday to develop a "people strategy" for the AI era, with state governments, philanthropic groups and employers ranging from Amazon.com and Microsoft to Bank of America and Eli Lilly, per the WSJ. Former Commerce Secretary Gina Raimondo, who served under President Joe Biden, and former Indiana Gov. Eric Holcomb, a Republican, will co-lead the effort. Business groups and lawmakers have called for more action to prepare the U.S. workforce for potential AI upheaval.

Read at Axios ↗ • Read at WSJ ↗

IV.Capability & Research Watch

Google DeepMind ships computer use as a built-in tool in Gemini 3.5 Flash, with two enterprise safeguards

Google DeepMind announced Wednesday that computer use is now a built-in tool in Gemini 3.5 Flash, previously available only as a standalone Gemini 2.5 model, allowing developers to build agents that act across browser, mobile and desktop environments. DeepMind said it used targeted adversarial training against prompt injection and released two optional enterprise safeguards: a requirement that users confirm sensitive or irreversible actions, and an automatic stop when an indirect prompt injection is detected. The company is offering the capability through the Gemini API and Gemini Enterprise Agent Platform and is encouraging developers to combine it with sandboxing, human-in-the-loop verification and strict access controls. Demos showed Gemini 3.5 Flash auditing its own documentation for accessibility issues and returning a categorized list of Gemini app features.

Read at DeepMind ↗

Stripe-led Intercept, backed by Anthropic and OpenAI Foundation, commits up to $500 million to stop respiratory viruses

Stripe is funding a new nonprofit called Intercept that will deploy up to $500 million in grants and investments to prevent respiratory infections, with backers including Anthropic, Flu Lab, the OpenAI Foundation, Bill Gates and traders at Jane Street Capital, per MIT Technology Review. The effort will fund vaccine work and large-scale air-cleaning systems for schools, offices and other public spaces, with an eventual goal of getting rid of respiratory viruses altogether. Stripe executive Nan Ransohoff and venture capitalist Charlie Petty are leading the initiative, modeled on Stripe's earlier $1.8 billion Frontier carbon-removal program. Intercept took shape after Ransohoff spoke with University of Washington structural biologist David Veesler, who said broad countermeasures against many viruses at once are technically possible.

Read at MIT Technology Review ↗

V.Industry & Market Watch

AI data center buildout is a "third wave" of U.S. inflation, WSJ columnist argues

The AI data center buildout has become a third wave of U.S. inflation, following tariffs and a fuel-price spike that are now receding, WSJ economics columnist Justin Lahart wrote Wednesday. In a National Association for Business Economics survey released Monday, 81% of economists said the buildout will add to inflation over the next year. Demand for memory chips is raising consumer-electronics prices, with Nintendo, Microsoft and Sony already charging more for devices and Apple's Tim Cook calling the cost jump unlike anything he had seen in over 40 years. Data centers are also straining the grid, with Goldman Sachs expecting them to drive nearly half of U.S. power demand growth through 2030.

Read at WSJ ↗

SK Hynix files for $29.65 billion U.S. listing, with trading slated for July 10

South Korea's SK Hynix, the world's second-largest memory chipmaker and a primary supplier of the high-bandwidth memory used in Nvidia AI accelerators, plans to raise about $29.65 billion through a U.S. listing by issuing American depositary receipts on Nasdaq, according to a Wednesday filing with the Korea Exchange, per the WSJ. The company expects trading to begin July 10. The deal would be one of the biggest share sales in history.

Read at WSJ ↗ • Read at Reuters ↗

Micron Q3 profit jumps roughly 15-fold and lifts AI chip stocks by $400 billion

Memory chipmaker Micron Technology reported a third-quarter profit roughly 15 times last year's level and forecast quarterly earnings above analyst estimates, sending its shares up 12% in extended trade and adding more than $400 billion in market value to a broader AI chip stock rally, per Reuters. Western Digital, Sandisk and Seagate Technology all jumped more than 8%, with Arm Holdings up about 6%, Marvell up almost 4%, and Broadcom up 2%. Micron has gained more than 260% year to date through Wednesday's close, while the PHLX chip index is up 90% for 2026.

Read at Financial Times ↗ • Read at Reuters ↗ • Read at Reuters ↗

Amazon commits another $13 billion to India for AI and cloud, bringing total to $48 billion through 2030

Amazon committed an additional $13 billion to its India AI and cloud infrastructure buildout through 2030, accelerating its expansion in the country, per Bloomberg. CEO Andy Jassy said the new spend brings Amazon's total India investment to roughly $48 billion through 2030, per Business Standard. The investment targets cloud infrastructure to deliver AI services in India.

Read at Bloomberg ↗ • Read at Business Standard ↗

VI.Global & Geopolitics

Hyundai's South Korean union approves strike over wage talks and AI rollout concerns

Hyundai Motor's South Korean union voted Wednesday to approve strike action, with members citing fears of robots replacing them, per the Financial Times. The union is seeking a 149,600 won ($96.82) increase in base monthly pay, a performance bonus equal to 30% of last year's net profit, and guarantees on employment and working conditions tied to the use of AI, per Reuters. The vote followed a collapse in wage talks and a National Labor Relations Commission mediation filing, with workers concerned that AI powered robots and growing overseas production could reduce overtime-tied earnings, per Korea JoongAng Daily. The union has roughly 39,000 members.

Read at Financial Times ↗ • Read at Reuters ↗ • Read at Korea JoongAng Daily ↗