AI Policy · Daily

Sen. Warren invoked a 2018 export control law to compel the Commerce Department to produce records on a China chip loophole she says the agency created by dropping Biden-era diffusion rules. Sens. Schiff and Banks introduced a bill giving U.S. AI firms an antitrust carve-out to share threat information and jointly delay high-risk model deployments against alleged Chinese copying. White House divided over how to respond to Chinese AI firm's alleged distillation of U.S. models. Beijing sent a senior diplomat to Washington to pin down the agenda for September AI talks ahead of President Xi Jinping's Sept. 24 U.S. visit.

I.Top Stories

Sen. Warren invokes 2018 statute to force BIS to hand over records on China chip loophole

Ranking Member Sen. Elizabeth Warren (D-Mass.) sent Under Secretary for Industry and Security Jeffrey Kessler a letter Thursday invoking the Export Control Reform Act of 2018 to compel the Bureau of Industry and Security, the Commerce arm that controls chip exports, to produce documents on export control loopholes she says BIS's own actions created. Warren says the Trump administration rescinded and stopped enforcing Biden's AI Diffusion Rule, stripping the underlying license requirement that made the January 2025 Foundry Due Diligence Rule effective. That could let foundries such as TSMC ship advanced AI chips to unvetted intermediaries in third countries, she wrote. "BIS has had every opportunity to fix this loophole and has yet to act," Warren wrote. She set an Aug. 6, 2026 deadline for BIS to produce the requested records.

Read at Senate Banking ↗

Schiff-Banks bill would grant U.S. AI firms antitrust exemption to coordinate on alleged Chinese distillation

Sens. Adam Schiff (D-Calif.) and Jim Banks (R-Ind.) and Reps. Bob Latta (R-Ohio) and George Whitesides (D-Calif.) introduced the Collaboration on Adversarial Threats and Security Risks Act on Thursday, per NBC News. The bill would let AI companies and industry groups share information on AI security risks in good faith and jointly delay or limit deployment of high-risk models with written notice, without running afoul of antitrust laws. It treats alleged Chinese AI copycats as a U.S. security threat and gives the Justice Department authority to seek an injunction against organizations that abuse the exemption. Sponsors likened the carve-out to the 2015 congressional exemption for cybersecurity information sharing. Latta, chair of the House Energy and Commerce subcommittee on energy, said AI models are now critical infrastructure alongside hospitals, water systems and the electric grid.

Read at NBC News ↗ Read at Senator Banks ↗

White House divided over how to respond to Chinese AI firm's alleged distillation of U.S. models

Trump administration officials are split over whether to punish Moonshot AI over allegations it copied American models to build Kimi K3, POLITICO reported. Treasury Secretary Scott Bessent and Office of Science and Technology Policy Director Michael Kratsios have pushed a hard line on intellectual property, while Commerce Secretary Howard Lutnick has taken a more modest approach centered on companies' access to cheaper, effective technology. Kratsios said on X Wednesday that the administration had information indicating Moonshot distilled Anthropic's Fable model and had obtained restricted Nvidia GB300 servers, but he provided no evidence, and the claims drew skepticism from researchers, industry and people around the administration — including over whether a frontier model could be copied in a matter of days. The post was not coordinated with the Commerce Department, which controls the Entity List and has not started drafting plans to add Moonshot or other Chinese AI companies; Commerce's Center for AI Standards and Innovation and the UK AI Security Institute jointly evaluated the model and found it still trails U.S. systems while ranking as the strongest open-weight model. Bessent, who is expected to lead the next round of U.S.-China economic talks, wrote that "open source is not open season on American IP" and said sanctions and Entity List designations remain possible. Anthropic has welcomed tighter scrutiny of Chinese developers; nearly 200 startup founders urged the administration this week not to cut off access to Chinese open-weight models. Trump has not weighed in publicly.

Read at POLITICO ↗

Beijing seeks U.S. clarity on September AI dialogue agenda ahead of Xi visit

Chinese Vice Foreign Minister Ma Zhaoxu traveled to Washington this week to press U.S. officials, including Deputy Secretary of State Landau, on the scope, possible outcomes and agenda of the planned September AI talks so Beijing can decide which officials to send, Bloomberg reported. Beijing's priority list for the discussions includes access to Anthropic's Mythos model, how the U.S. plans to control the release of future frontier models, and whether Washington will impose restrictions on Chinese open weight AI models, analysts told Reuters. Treasury Secretary Scott Bessent will lead the U.S. side, and the meeting is expected before Chinese President Xi Jinping's Sept. 24 U.S. visit. White House press secretary Karoline Leavitt said Thursday that AI is a priority for the president and that the administration wants to ensure the U.S. continues to lead in the AI race.

Read at Bloomberg ↗ Read at Mint ↗

Bipartisan House bill would require audits, incident reporting from largest AI developers

Reps. Lori Trahan (D-Mass.) and Jay Obernolte (R-Calif.) introduced the FRONTIER Act on Thursday, a national risk-based oversight framework carved out of their broader Great American AI Act discussion draft. Covered developers would have to publish model cards and transparency reports at deployment, maintain risk-management frameworks reviewed annually or within 30 days of any material change, report critical safety incidents within 24 hours of discovery, and submit to third-party independent audits. The Washington Examiner reported that coverage is limited to developers that have spent more than $1 billion on model development over the past three years; the bill would preempt state laws within the scope of its provisions. Original cosponsor Erin Houchin's (R-Ind.) statement pointed to an advanced AI system that "broke out of its own developer's testing environment" this week, referencing the OpenAI model that escaped containment and hacked Hugging Face without human direction.

Read at Washington Examiner ↗ Read at Trahan ↗ Read at Obernolte ↗

OSTP Director Kratsios endorses more resources for CAISI at House Science hearing

Michael Kratsios, director of the White House Office of Science and Technology Policy, told Rep. Jay Obernolte (R-Calif.) at a House Science Committee hearing Wednesday that "CAISI can certainly benefit from more resources," giving administration cover to Obernolte's bipartisan bill (H.R. 9363) to codify the Commerce Department's AI safety body into law, per FedScoop. Obernolte said he believes the Center for AI Standards and Innovation likely needs "closer to $100 million a year"; the bill currently provides $20 million annually and CAISI received $10 million in fiscal 2026, the same as each of the prior two fiscal years. Kratsios said the office needs resources to recruit high-talent staff for temporary government tours to do frontier model testing and evaluation, work that pays far better in the private sector. The House Science panel unanimously advanced the codification bill last month as part of an AI package.

Read at FedScoop ↗ Read at House Science ↗

II.China Watch

Microsoft weighs shifting some Copilot workloads to Moonshot's Kimi K3

Microsoft is internally testing Moonshot AI's Kimi K3 to see whether Copilot inference requests currently handled by OpenAI and Anthropic models could be moved to the Chinese open weight model, TechNode reported. Microsoft has not decided to replace any model but estimates the shift could cut its annual cloud infrastructure bill by up to $600 million, the report said. The evaluation is part of a broader push to route requests across multiple models to control the cost of serving high volumes, though any deployment would still have to clear data sovereignty, security and export control questions and would likely begin with less sensitive workloads. The interest gives Moonshot's Kimi K3, released July 16, its first plausible route into a major U.S. cloud stack and adds commercial weight to the pitch Beijing made at the World AI Conference that Chinese open source AI can serve as a global standard.

Read at TechNode ↗

Pentagon puts Fudan and Shanghai Jiao Tong on research restriction list

The Pentagon has added Fudan University and Shanghai Jiao Tong University to a list of foreign institutions deemed a threat to national security, barring U.S. Defense Department funded researchers from collaborating with them or using their equipment starting in the 2026 fiscal year, per SCMP. The updated list identifies 88 institutions in mainland China. It had previously focused mainly on People's Liberation Army academies and universities tied to China's defense industry; the shift to two elite civilian campuses widens the perimeter into schools with dense international science ties. Fudan and Shanghai Jiao Tong lead China in AI, materials and biotech research and have long trained talent that moves between American and Chinese labs. The listing pushes Washington's research security regime well past defense linked institutions and would sever a slice of the federally funded U.S.-China joint programs that have long channeled top Chinese researchers into U.S. labs.

Read at SCMP ↗

Intel and AMD sign longer server CPU deals with Chinese buyers as prices climb

Intel and Advanced Micro Devices are signing longer-term purchase commitments with Chinese server customers for data center processors as prices surge, Reuters reported, citing two people familiar with the talks. The agreements typically lock in volumes but not prices; most cover about a year of supply, though the companies have discussed commitments of two years or longer with some customers, one of the sources said. Server CPU prices in China are up more than 10% month on month for some products and more than 40% since the start of the year, that source said. The squeeze echoes the AI-driven memory chip shortage, which pushed buyers toward long-term supply deals, and could raise costs and slow deployment for Chinese cloud providers and internet companies expanding AI services. Chinese buyers are competing for Intel and AMD processors even as separate U.S. restrictions limit their access to the most advanced AI GPUs.

Read at Reuters ↗

Nearly 100 Chinese companies compete to supply embodied AI training data

Nearly 100 Chinese firms now compete to produce training data for embodied AI, and 15 independent data service providers raised about 4.47 billion yuan (roughly $623 million) in the 12 months through July, per QbitAI. Of the 97 tracked players, 70 do data collection and 27 build data infrastructure, and 43% of the collection firms mix methods that include teleoperation on real robots, human motion capture, simulation and internet video distillation. Lightwheel Intelligence dominates the top tier with about 3.1 billion yuan (about $430 million) across six funding rounds and a valuation above $2 billion, described in the report as the world's first embodied data unicorn. Data collection factories now sit in 20 provinces, concentrated in the Yangtze River Delta, and state backed platforms run by local Chinese governments account for roughly a quarter of the tracked players in a sector Beijing has designated a national priority industry.

Read at QbitAI ↗

III.Policy Tracker

House Energy and Commerce clears Ratepayer Protection Act 52-0 as Senate companion introduced

The House Energy and Commerce Committee advanced the bipartisan Ratepayer Protection Act, H.R. 9340, on Tuesday in a unanimous 52-0 vote, per The Hill. The bill, sponsored by Reps. Gabe Evans (R-Colo.) and Kathy Castor (D-Fla.), would require states to consider federal standards directing data center customers to pay the full cost of new generation and transmission needed to serve them. It codifies the White House Ratepayer Protection Pledge that Trump extended to state governors and utilities including NextEra and Duke Energy, as covered in AIPD's July 23 edition. Sen. Jon Husted (R-Ohio) introduced a Senate companion, S. 5028, last week. Chair Brett Guthrie (R-Ky.) has predicted a floor vote and made the bill a committee priority ahead of the summer recess.

Read at The Hill ↗ Read at E&E News ↗

Amazon requires sellers to label AI generated people in ad images under New York synthetic performer law

Amazon has begun requiring third-party marketplace sellers to label AI generated people in product images, following a New York statute that mandates disclosure when an ad uses a "synthetic performer" in place of a human actor, per CNBC. Gov. Kathy Hochul signed the law in December 2025; it went into effect June 9 and carries penalties of $1,000 for a first violation and $5,000 for each further violation. The statute defines synthetic performers as digitally created media that appear as a real person and carves out movies, TV shows, streaming content and video games. It was opposed by the American Association of Advertising Agencies and backed by actors' union SAG-AFTRA.

Read at CNBC ↗ Read at The Washington Post ↗

Oregon separates state chief AI officer role from privacy chief, names Nevada hire as new privacy lead

Oregon CIO Terrence Woods appointed Michael Hanna-Butros Meyering, previously chief communications and policy officer at Nevada's Governor's Technology Office, as Oregon's chief privacy and communications officer, per StateScoop. Meyering succeeds Nik Blosser, who had served as Oregon's first chief privacy officer and AI strategist since August 2025 and will now serve as the state's chief AI officer. Woods said the split gives each area dedicated leadership as both become more important to how the state governs data, procures technology and reviews AI use cases.

Read at StateScoop ↗ Read at Oregon Department of Administrative Services ↗

IV.Capability & Research Watch

AMD CEO Lisa Su defends open source AI at Advancing AI conference, echoing Nvidia's Huang

AMD CEO Lisa Su said "I think open source is a great thing" at a press event at the company's Advancing AI conference in San Francisco on Thursday, per Fortune. Su said restricting open models needs careful management rather than an outright ban, pushing back on White House deliberations over restrictions on foreign open source AI. Her remarks come days after Nvidia CEO Jensen Huang told Axios that Chinese open source models are "excellent" and should be embraced, as covered in AIPD's July 23 edition. AMD executives pointed to industry-led self-regulation via open constitutions and guardrails as an alternative to government rules, without providing specifics.

Read at Fortune ↗

OpenAI launches Health in ChatGPT one day after California lawsuit seeks to halt it

OpenAI began rolling out Health in ChatGPT to U.S. users 18 and older on Thursday across its free and paid tiers, letting people link medical records from Apple Health, One Medical and supported hospital systems and ask the chatbot to compare lab results, summarize appointment changes and track sleep against activity, per SiliconANGLE. Former Florida pastor Scott Winters sued OpenAI and CEO Sam Altman in San Francisco Superior Court on July 22, alleging ChatGPT-4o's medical advice caused him to delay treatment for symptoms that later produced a near-fatal pulmonary embolism. The suit brings claims of strict liability, negligence, invasion of privacy and practicing medicine without a license, and asks the court to halt Health in ChatGPT until independent auditors verify its safety. OpenAI paired the launch with a caveat that ChatGPT can still get medical questions wrong.

Read at SiliconANGLE ↗ Read at OpenAI ↗

V.Industry & Market Watch

Oracle signs $7B, 10-year Pentagon software deal covering every service and the intelligence community

Oracle signed a nearly $7 billion, decade-long contract to supply software licenses, maintenance and consulting to every U.S. military branch, the Coast Guard, and the intelligence community, per SiliconANGLE. The five-year base period covers perpetual and subscription software licenses, with option years extending the deal to 10 years. Defense Department CIO Kirsten Davies said the deal will save taxpayers at least $441 million by consolidating how the Pentagon procures on-premises Oracle capabilities. The award is part of Defense Secretary Pete Hegseth's push to cut duplicate DoD software contracts and expands Oracle's federal footprint alongside its role in the $500 billion Stargate AI data center initiative.

Read at SiliconANGLE ↗

Nvidia commits $1.5B to Amkor for U.S. advanced packaging expansion

Nvidia entered a multi-year agreement with Amkor Technology on Thursday valued at $1.5 billion to expand advanced semiconductor packaging and test capacity in the United States, including in Arizona. Nvidia will make a prepayment to support Amkor's U.S. operations, and the companies will jointly develop packaging and testing technology for AI and accelerated computing platforms that combine different chip types in a single package. Amkor already supplies advanced packaging for Nvidia's data center processors, entered a 10-year partnership with TSMC in June, and is also working with AMD. Amkor shares jumped 17% in extended trading.

Read at Bloomberg ↗ Read at Reuters ↗

AMD unveils Helios rack and MI455X accelerator; OpenAI to deploy Helios starting this year

AMD's Helios server rack, with up to 72 MI455X AI accelerators paired with the new Venice CPU, is in full production and will begin shipping at the end of the third quarter, the company said Thursday, per Investing.com. OpenAI's vice president of compute strategy Sachin Katti joined CEO Lisa Su on stage in San Francisco to say OpenAI plans to deploy Helios extensively starting late this year and accelerating through 2027. OpenAI also expects to use AMD's next-generation MI500 chips. Cerebras CEO Andrew Feldman also joined to announce a Cerebras-AMD partnership offering combined AI servers, first in Cerebras data centers. AMD projects a $2 trillion global computing market by 2030, with $1.4 trillion from AI accelerators and $220 billion from CPUs; the company estimated the total market at $365 billion in 2025.

Read at Bloomberg ↗ Read at Investing.com ↗

Anthropic and OpenAI push free and cut-price offerings into a claimed $6T education market

Anthropic and OpenAI are moving into a global education market that vendors estimate at about $6 trillion, offering free and cut-price tools to educators and students, per Financial Times. Anthropic launched Claude for Teachers on July 14, giving K-12 U.S. educators free premium access to Claude Cowork and Claude Code and saying a dedicated schools and districts offering is next. OpenAI has operated a similar ChatGPT for Teachers program since November for eligible U.S. educators.

Read at Financial Times ↗ Read at 9to5Mac ↗

Centrica cuts 1,300 UK call center jobs as customer calls fall 20%

Centrica, the FTSE 100 owner of British Gas, will cut 1,300 call center jobs over two years, including 800 tied to an AI tool deployment on top of 500 announced last month, a 14% reduction across customer-service teams in six UK cities, per The Guardian. CEO Chris O'Shea said "AI isn't driving these particular job reductions" but that they reflect a 20% drop in customer calls, with more than 90% of customers using digital channels first. Centrica reported first-half retail profits of £346 million (about $450 million), up from £338 million (about $440 million) in the same period last year.

Read at The Guardian ↗

VI.Global & Geopolitics

APEC digital and AI ministers in Chengdu adopt joint declaration backing open source cooperation with security guardrails

Ministers from Asia-Pacific economies adopted a joint declaration on digital and AI cooperation at the 2026 APEC Digital and AI Ministerial Meeting in Chengdu on Thursday, per CNBC. China's Minister of Industry and Information Technology Li Lecheng, who chaired the session, said the statement was the first APEC ministerial to include open source cooperation at the minister level, with the U.S. and other economies endorsing security guardrails for open source AI. The declaration frames cooperation on standards setting, industrial development and infrastructure across the 21 APEC economies as a framework for regional digital and AI cooperation over the coming years, per People's Daily Online.

Read at CNBC ↗ Read at People's Daily Online ↗ Read at APEC ↗

U.S. joint venture to produce Ukraine's Magura autonomous sea drones for first time

Ukrainian defense company Uforce and Oregon boatyard ReconCraft signed an agreement to build Ukraine's combat-tested Magura autonomous surface vessels in the United States, per Bloomberg and The Maritime Executive. The 24-foot Magura drones, credited with securing sea denial against Russian shipping in the Black Sea, will be produced at prices below $500,000 per unit, a fraction of the cost of a small antiship missile. Sean Plankey, CEO of Uforce's U.S. operations and a former senior adviser to then-Homeland Security Secretary Kristi Noem who helped drive the Coast Guard's Force Design 2028 program, will lead the U.S. effort. The partnership targets high-volume U.S. Navy procurement contracts as the White House pushes to expand America's unmanned arsenal.

Read at Bloomberg ↗ Read at The Maritime Executive ↗