AI Policy · Daily

U.S. and European officials opened talks on a "trusted partner" scheme that would let allies test American frontier AI models. A coalition of state legislators urged Congress to strip the three-year preemption clause from the draft Great American AI Act (GAAIA). Lawmakers in both parties see fresh momentum for AI guardrail legislation after the administration's move against Anthropic, though election-year consensus remains elusive. Anthropic's confidential IPO filing approaches a $1 trillion valuation as investors weigh the risk of a federal shutdown of its Mythos and Fable models. G7 leaders close their summit in France Wednesday with an AI working lunch featuring the chiefs of OpenAI, Anthropic and Google DeepMind.

I.Top Stories

U.S. and Europe begin discussing 'trusted partner' scheme for allied access to frontier AI models

U.S. and European officials began discussing a "trusted partner" scheme that would grant Washington's close allies priority access to U.S. frontier AI models, the Financial Times reported. Commerce Secretary Howard Lutnick floated the concept with European diplomats on the sidelines of the G7 summit in France. The proposal responds to the Commerce Department's export action against Anthropic's Mythos 5 and Fable 5. European Commission Executive Vice-President Henna Virkkunen urged Washington to clarify its security concerns and warned against "discriminatory" measures targeting partners such as the EU, per Firstpost. The Software and Information Industry Association, whose members include Apple, Amazon and Google, separately warned that "arbitrary use of discretionary authority" over frontier models risked undermining the global spread of American AI.

Read at Financial Times ↗ • Read at Firstpost ↗

State lawmaker coalition urges Congress to drop Great American AI Act's three-year preemption clause

A coalition of state lawmakers submitted a letter to Congress on Tuesday opposing the Great American AI Act, the federal AI bill released in draft form earlier this month, StateScoop reported. The letter targets GAAIA's proposed three-year preemption of state AI laws that regulate the development of AI models. Americans for Responsible Innovation, the policy nonprofit that assembled the letter, said 51 of Tuesday's signatories also signed a similar letter the group organized last June. At that time, more than 260 state legislators objected to a 10-year ban tucked into the federal budget reconciliation bill. That earlier moratorium was dropped after a 99-1 Senate vote, before Republican leaders began floating standalone or alternative vehicles for the preemption.

Read at StateScoop ↗

White House move against Anthropic revives congressional push for AI guardrails

The Trump administration's moves to rein in Anthropic are giving fresh momentum to efforts in Congress to set guardrails on frontier AI models, Politico reported. In roughly a dozen interviews this week, lawmakers said they were caught off guard by Friday's export ban on Fable 5 and Mythos 5 and had yet to receive a formal briefing; Senate Commerce Chair Ted Cruz (R-Texas) said he had seen only "what's been reported in the press." Members of both parties said they see an opening to reclaim congressional authority over AI, though consensus remains elusive in an election year, with Democrats favoring mandatory review requirements and Republicans wary of a heavy hand. Sen. Todd Young (R-Ind.) said the episode could "conceivably" push lawmakers toward legislation establishing a system to review new models, warning against "just trusting private actors to do the right thing."

Read at Politico ↗

Anthropic's confidential $1 trillion IPO pitch now carries explicit federal shutdown risk

Anthropic has filed confidentially to go public as soon as this fall at a valuation approaching $1 trillion, Fortune reported, with investors weighing whether the price accounts for the risk that the government switches off its flagship product overnight. The offering grows more complicated with Fable 5 off the market, Semafor reported, as staff flocked to Washington this week to triage with the officials who imposed the export controls Friday. Anthropic must balance its safety-first reputation against getting along with the government, with Semafor's Reed Albergotti writing that "no AI model has ever avoided being jailbroken."

Read at Fortune ↗ • Read at Semafor ↗

OpenAI says ChatGPT will deploy on Pentagon's GenAI.mil in early July

OpenAI will bring ChatGPT to the Pentagon's GenAI.mil generative AI platform in early July, Mohammed Husain, the company's strategic delivery lead for cyber, said at the Defense One Tech Summit in Arlington, Virginia. The deployment will make ChatGPT available to more than 3 million defense personnel for sensitive-but-unclassified work, with certification at Impact Level 5. The Pentagon launched GenAI.mil in December with Gemini for Government, and OpenAI and xAI models followed. As of late April, the platform had more than 1.3 million regular users and more than 100,000 AI agents.

Read at Defense One ↗ • Read at OpenAI ↗

Washington Post investigation pins Anthropic's trust collapse on undisclosed Mythos recipient list

Anthropic gave the Trump administration a list of 111 organizations slated to receive advanced Mythos access several weeks ago, the Washington Post reported, citing two White House officials. The company later disclosed that the list had ballooned by roughly 50 additional entities. Senior administration officials began weighing export controls after the company did not identify the new recipients for days. The trust dispute predated the public takedown order by weeks. The Financial Times editorial board separately called the administration's approach to AI policymaking "arbitrary" in a piece published Tuesday.

Read at Washington Post ↗ • Read at Financial Times ↗

II.China Watch

Tencent's WeChat Pay opens its rails to AI agent purchases as Chinese super apps race to host AI commerce

WeChat Pay on Wednesday integrated with Tencent's in-house desktop agent WorkBuddy to let AI agents initiate and complete payments inside the WeChat app, per Caixin. The integration lets a user ask the agent for a purchase. The agent then recommends, orders and triggers payment, with the user confirming each deduction on their phone. Tencent plans to open the AI payment capability to additional agent platforms beyond WorkBuddy.

Read at Caixin ↗

Inland Chinese cities take a bigger share of the country's AI computing buildout

Five small cities are absorbing a growing share of China's AI compute infrastructure: Shaoguan in Guangdong, Shantou, Zhongwei in Ningxia, Qingyang in Gansu and Danzhou in Hainan, Pandaily reported. Shaoguan was approved in January as China's only pilot cluster for processing overseas data, with dedicated high-bandwidth fiber lines handling Hong Kong and Macau traffic. Qingyang has converted oil and coal land into more than 150,000 PFLOPS of compute capacity serving inference workloads for Southeast Asian customers of major Chinese internet platforms. Zhongwei, where Amazon built its first China data center cluster, now hosts six major cloud platforms serving more than 4,000 enterprises and uses its cool climate for 280 days a year of free air cooling, 40% more energy efficient than eastern cities. The dispersion concentrates Chinese AI compute outside the coastal cities most exposed to U.S. export controls on advanced accelerators, giving Beijing a wider domestic footprint to allocate during chip shortages.

Read at Pandaily ↗

Former Alibaba Qwen chief Lin Junyang closes initial round for new AI lab at $2 billion valuation

Lin Junyang, the former head of Alibaba's Qwen large language model team, closed an initial round of several hundred million dollars for his new AI lab Pragmatics at a post-money valuation of about $2 billion, The Information reported. Gaorong Capital and Sequoia China co-led with $100 million each, and Tencent followed with $20 million. Lin registered Shanghai Pragmatics Technology and a related Shanghai entity called Yuyong Technology in May and June, per Qichacha business records. Two people familiar with the matter said Lin is already in talks for a follow-on round. The investment puts Tencent behind a venture that may compete with its own AI program at a time when U.S. export controls keep frontier compute scarce in China.

Read at The Information ↗

China publishes AI safety section in global governance white paper as G7 wraps in Evian

China's State Council Information Office on Wednesday released a global governance white paper titled "More Just and Equitable Global Governance: China's Principles, Proposals and Actions," which includes a section on AI safety, CNBC reported. The release coincided with the G7's closing AI session at Evian-les-Bains, where Beijing is absent. Foreign Minister Wang Yi said at a press conference marking the white paper that China is accelerating preparations for the World Artificial Intelligence Cooperation Organization and welcomes all parties to join, per China Daily Asia. Zhou Haibing, deputy head of the National Development and Reform Commission, said China will host the 2026 World Artificial Intelligence Conference and High-Level Meeting on Global AI Governance in Shanghai in July.

Read at CNBC ↗ • Read at China Daily Asia ↗ • Read at State Council of China ↗

III.Federal Policy Tracker

SAG-AFTRA letter backing the NO FAKES Act reaches 16,000 signatures ahead of Thursday Senate Judiciary markup

SAG-AFTRA's open letter calling on Congress to pass the bipartisan NO FAKES Act has gathered more than 16,000 signatures, TheWrap reported. The bill would give every individual, including nonpublic figures, control over how their name, voice and likeness are used and establish a federal cause of action against unauthorized deepfakes. The bill was revived last month after it stalled in committee last year. All four Senate co-sponsors sit on the Senate Judiciary Committee, which will address the bill on Thursday. AFL-CIO, YouTube, IBM, OpenAI, the Motion Picture Association, The Recording Academy, the National Association of Broadcasters and the Recording Industry Association of America have also endorsed the bill.

Read at TheWrap ↗

Sen. Rick Scott introduces bill mandating public terrorism risk reports on generative AI

Sen. Rick Scott (R-Fla.) on Monday proposed the "Generative AI Terrorism Risk Assessment Act," which would require the federal government to issue regular public reports assessing how terror organizations could be using AI tools, per his office. Scott wrote on X that "AI is creating a world of possibilities; it's also creating new National Security threats."

Read at Sen. Rick Scott ↗ • Read at Benzinga ↗

Local Pennsylvania backlash over AI data center buildout sets up swing state political fight

A proposed methane powered data center campus at the former Pennhurst asylum site near Philadelphia would span nearly 2 million square feet across three buildings and sit less than 600 feet from a state veterans' home, per The Guardian. Donald Hyman, a 62-year-old veteran at the Southeastern Veterans' Center, told the publication that air pollution from the on-site power plant and backup generators could disrupt his recovery from congestive heart failure. He and four other veterans at the home said they fear the noise could trigger residents with post-traumatic stress disorder. The Pennhurst project sits in a state where Gov. Josh Shapiro and state lawmakers are publicly clashing over the data center facilities powering an AI compute buildout. "You're trying to force something on us we don't want. We don't want it, period," Hyman said.

Read at The Guardian ↗

IV.Industry & Market Watch

SpaceX agrees to $60 billion stock acquisition of AI coding startup Cursor

SpaceX agreed Tuesday to acquire AI coding startup Cursor in a $60 billion stock deal expected to close in the third quarter, per TechCrunch. The transaction exercises an April option in which SpaceX could either acquire Cursor for $60 billion in stock or pay a $10 billion breakup fee. Cursor was on track to close a separate $2 billion fundraising round at a $50 billion valuation led by Andreessen Horowitz, Thrive and Nvidia before SpaceX moved. The deal folds Cursor into SpaceX's AI division, built around xAI, which SpaceX merged with earlier this year.

Read at TechCrunch ↗ • Read at AP ↗ • Read at CNBC ↗

Mobileye plans U.S. robotaxi launch in 2027 with 100-vehicle initial fleet

Mobileye said Tuesday it will launch its own robotaxi service in a U.S. city in 2027, starting with 100 autonomous vehicles and scaling to roughly 17,000 over the following five years, per TechCrunch. The move puts the Intel subsidiary in direct competition with robotaxi operators such as Waymo, Zoox and Tesla, even as it continues to supply its self-driving system to automakers including Volkswagen and its MOIA unit. CEO Amnon Shashua said in a statement that operating the service will let the company accelerate adoption, gain direct operational experience and demonstrate the technology, per Reuters.

Read at TechCrunch ↗ • Read at Reuters ↗

Nvidia CEO Huang calls for 'new social norms' as Coherent breaks ground on $50 million CHIPS funded Texas optical fab

Nvidia CEO Jensen Huang told the Associated Press on Tuesday that society needs "new social norms" as AI becomes pervasive. Huang spoke at the groundbreaking of Coherent's expanded indium-phosphide facility in Sherman, Texas. Coherent will receive a $50 million CHIPS Act grant for the Sherman expansion, on top of roughly $17 million in earlier support from the Texas CHIPS program and the Sherman Economic Development Corporation, per Nvidia. The plant produces indium-phosphide wafers that carry data among chips at the speed of light, the optical backbone of modern AI infrastructure. The expansion is part of Nvidia's commitment to bring up to $500 billion of AI infrastructure to U.S. soil through industry partnerships with new sites in Arizona and Texas.

Read at AP ↗ • Read at Nvidia ↗

V.Global & Geopolitics

UK details ID-or-face-scan enforcement as it explores VPN limits to police under-16 social media ban

Anyone opening a new social media account in the UK will have to prove they are over 16 by uploading an ID or passing a facial age scan from spring 2027, BleepingComputer reported. Long-standing accounts are largely exempt, but signing up fresh now triggers verification, effectively ending anonymous account creation in the UK. Security and privacy experts warned the checks are easy to circumvent and put users' ID and biometric data at risk of breaches. The UK government is also examining steps to stop children from circumventing the ban via VPNs, per the Financial Times. The ban covers Instagram, YouTube, TikTok, Snapchat, Facebook and X. WhatsApp, Signal and YouTube Kids are excluded, with a narrow exemption list for educational services, e-commerce and music streaming.

Read at BleepingComputer ↗ • Read at Financial Times ↗ • Read at GOV.UK ↗

Google DeepMind partners with UK government on AI accelerated planning prototype

Google DeepMind, the UK government, Google Cloud and consulting firm Faculty are co-developing an AI prototype to halve the time it takes for local councils to process homeowner planning applications, the company said. The tool, built with Gemini, is undergoing trials in Barnet, Camden and Dorset, with national rollout planned from 2027. Householder applications account for nearly 70% of UK planning applications each year. The prototype will pre-process backlogs, identify relevant national and local policies, summarize consultation feedback and draft first-cut assessment reports for officers to verify.

Read at DeepMind ↗

Bank of Korea warns AI driven semiconductor bonuses could stoke broader inflation

South Korea's AI driven semiconductor boom may complicate the inflation outlook as soaring bonuses at major technology firms fuel broader wage growth and consumer demand, Bloomberg reported, citing Bank of Korea Governor Shin Hyun-song. Samsung Electronics agreed to pay 10.5% of operating profit as a special performance bonus to employees in its semiconductor division over the next 10 years, per Chosunbiz. Shin said the structure "could create upward pressure on prices as wages boost purchasing power and increase demand." The Bank of Korea held its benchmark rate at 2.5% at its May 28 monetary policy meeting and raised its 2026 growth forecast to 2.6% from 2.0%.

Read at Bloomberg ↗ • Read at Chosunbiz ↗