AI Policy · Daily

Senate Banking Chairman Tim Scott convened an AI hearing that turned into a bipartisan push for AI chip export controls. Warren and Sen. Richard Blumenthal introduced the AI Bubble Transparency Act, directing Treasury's Office of Financial Research to gather AI exposure data from financial firms to flag systemic risks. Florida Gov. Ron DeSantis broke with the White House over its push to override state AI laws, calling federal preemption bad policy after the push killed his own Florida bill. Texas Gov. Greg Abbott ordered state utility regulators to shield ratepayers from data center power and water costs, with a July 17 deadline for a memorandum and 2027 priorities including repeal of the sales tax exemption.

I.Top Stories

Senate Banking's "AI and the American Dream" hearing turns into bipartisan push to advance chip export controls

Senate Banking Committee Chairman Tim Scott (R-S.C.) on Thursday led the panel's hearing "AI and the American Dream: Promoting Innovation, Affordability, and American Dominance." Ranking Member Sen. Elizabeth Warren (D-Mass.) said "Congress can no longer be a bystander as the risks from AI grow nearer," and pressed witnesses on advanced chips flowing to China and debt-financed AI buildouts. Warren called on Scott to mark up a slate of bipartisan bills aimed at keeping advanced AI chips out of China — the AI OVERWATCH Act, Chip Security Act and MATCH Act — saying, "I think we should mark these up this month." Sen. Pete Ricketts (R-Neb.) separately urged a markup of the MATCH Act, which he introduced with Sen. Andy Kim (D-N.J.) in April to restrict exports of chip manufacturing equipment to adversaries including China, arguing that "if you are going to control one part of the semiconductor supply chain, it should be the [semiconductor manufacturing equipment]."

Read at Politico Pro ↗ • Read at Senate Banking ↗ • Read at Senate Banking ↗

Warren and Blumenthal introduce AI Bubble Transparency Act directing OFR to collect AI exposure data from financial firms

Sen. Elizabeth Warren (D-Mass.) and Sen. Richard Blumenthal (D-Conn.) on Thursday introduced the AI Bubble Transparency Act, which would direct Treasury's Office of Financial Research to collect data from financial companies on their AI related exposures. Warren said the bill would give regulators and Congress what they need to identify systemic AI risks early and head off another preventable financial crisis. Blumenthal said the legislation targets the "opaque debt markets and shadowy circular arrangements" AI companies are using to fund their buildouts. The bill is endorsed by Americans for Financial Reform.

Read at Senate Banking ↗

DeSantis publicly opposes White House push to preempt state AI laws

Florida Gov. Ron DeSantis on Thursday characterized potential White House preemption of state AI laws as "bad policy and even worse politics," Politico reported. The Trump administration's opposition to state AI rules killed Florida AI legislation DeSantis had ranked among his top priorities for his final year in office. DeSantis warned that state-level AI safeguards in other capitals could be swept aside if federal preemption advances. Florida, led by DeSantis and state Attorney General James Uthmeier, is applying more pressure than any other Republican-led state on AI regulation, per Politico.

Read at Politico ↗

Texas Gov. Abbott directs PUCT and ERCOT to take immediate steps on data center infrastructure costs

Texas Gov. Greg Abbott on Wednesday directed the Texas Public Utility Commission (PUCT) and the Electric Reliability Council of Texas (ERCOT) to take immediate steps to ensure data centers do not shift their power and water costs onto Texas ratepayers, The Hill reported. Abbott's letter lays out 2027 legislative priorities including closed-loop water systems that reuse water over years, annual electricity and water reporting, and a repeal of the state's data center sales tax exemption, per the Texas Tribune. The Tribune estimates the existing exemption will cost Texas $3.2 billion in sales tax revenue over the next two years. KERA News reported the PUCT and ERCOT must submit a memorandum by July 17 outlining steps taken, statutory limits to their authority, and laws needed to act.

Read at The Texas Tribune ↗ • Read at KERA News ↗ • Read at The Hill ↗

Anthropic, OpenAI and Google executives plan to attend G7 leaders summit in France next week

Senior AI executives from Anthropic PBC, OpenAI and Alphabet's Google plan to attend the Group of Seven summit in Évian-les-Bains, France, June 15 to 17, Bloomberg reported. AI is set to feature on the agenda alongside the electricity grid pressures and online child protection topics G7 digital ministers flagged in their May Paris statement, per AFP. OpenAI's Sam Altman, invited by French President Emmanuel Macron, will attend for the first time. President Trump is also confirmed to attend the summit.

Read at Bloomberg ↗ • Read at Le Parisien ↗ • Read at The Economic Times ↗

II.China Watch

Turing laureates open Beijing's BAAI conference by calling AGI alignment unsolved

The eighth Beijing Academy of Artificial Intelligence conference opened June 12 with keynote speeches from Turing Award laureates Whitfield Diffie and Andrew Barto, both arguing that AGI safety is blocked by unsolved theoretical problems, Pandaily reported. Diffie said public key cryptography succeeded only after decades of formal specifications and standards work, and AGI cannot follow that path because no one can yet write a formal definition of behavior as basic as "not hallucinating." Barto, who shared the 2024 Turing Award for reinforcement learning, said designing reward functions for real world tasks is intractable, invoking Norbert Wiener's "Midas Touch" warning.

Read at Pandaily ↗

Chinese scholar calls EU tech sovereignty an "illusion" and urges Beijing to court Europe

The European Union's Technological Sovereignty Package, rolled out last week, is unlikely to deliver AI independence because frontier capability sits in the U.S. and China, a Chinese expert told the South China Morning Post. The expert urged Beijing to use friction between Brussels and the Trump administration to make Chinese AI products "indispensable" to middle powers, the report said. The EU package bundles a Chips Act 2.0, a Cloud and AI Development Act, an open source strategy and an energy digitalization roadmap.

Read at SCMP ↗

Nvidia's Latin America chief denies the region is a corridor for restricted GPUs into China

Marcio Aguiar, Nvidia's enterprise sales head for Latin America, told Web Summit Rio that the region has not served as a back door for sanctioned chips into China, but acknowledged the pressure from U.S. export controls is significant enough to reach his sales desk, the South China Morning Post reported. Aguiar said Nvidia walks away from orders when buyers in countries the company has never dealt with cannot explain end use or data center location. The remarks respond to an Anthropic paper published in mid-May that alleged Chinese labs relied partly on smuggled processors.

Read at SCMP ↗

All-China Federation of Trade Unions calls for new worker rights rules under AI

Workers' Daily, the All-China Federation of Trade Unions newspaper, closed a four-part series in Thursday's edition with an editorial urging government agencies to respond to AI risks for Chinese workers, Sinocism reported. The series said the rights protection issues of greatest concern to workers cluster around several areas, including the development of digital avatars of employees and the institutionalization of "AI skill packs" tied to performance evaluations. All four installments and the editorial were translated in full.

Read at Sinocism ↗

III.Federal Policy Tracker

AI Policy Network coalition urges Senate Banking to move bipartisan MATCH Act on multilateral AI chip controls

AI Policy Network President Mark Beall and 14 co-signers on Thursday sent a letter to Sens. Pete Ricketts (R-Neb.) and Andy Kim (D-N.J.) urging the Senate Banking Committee to advance S.4281, the Multilateral Alignment of Technology Controls on Hardware Act. The MATCH Act would mandate negotiations to align allied export controls on semiconductor manufacturing equipment with U.S. rules and close servicing and entity-specific loopholes, the coalition said. The letter argues that American firms alone bear the cost of restraint while China's chipmaking ambitions advance anyway, putting U.S. companies at a commercial disadvantage.

Read at AIPN ↗

OCC and Fed press banks on AI use during routine exams

The Office of the Comptroller of the Currency and the Federal Reserve have begun asking banks in routine examinations to map how they deploy AI in higher-risk areas such as lending, know-your-customer checks and sanctions screening, Reuters reported Friday, citing three people familiar with the matter. Supervisors are probing vendor and subcontractor risk, data safeguards, human oversight and "kill switches" that let firms shut systems down, while stopping short of AI-specific rules and leaning on existing model-risk and third-party oversight frameworks, the sources said. The scrutiny follows the OCC's April announcement that it, the Fed and the Federal Deposit Insurance Corporation plan a formal request for information on banks' use of generative and agentic AI.

Read at Reuters ↗

Technology Modernization Fund head says VA AI forms project will save 2,000 years of veterans' time, asks Congress for fiscal 2027 reauthorization

Technology Modernization Fund acting director Jessie Posilkin told Federal News Network's Cloud Exchange on Wednesday that the fund's $7.4 million AI investment to digitize and centralize VA forms will save more than 2,000 years of veterans' time, FedScoop reported. Posilkin said the TMF has more than $200 million left for similar agency projects but needs congressional reauthorization in fiscal 2027 to expand the pipeline. Posilkin said the scale and scope of government technology challenges keeps rising with AI adoption, making the remaining funds a drop in the bucket.

Read at FedScoop ↗

IV.Capability & Research Watch

Google DeepMind announces $10 million multi-agent AI safety research pot with Schmidt Sciences, UK ARIA and Cooperative AI

Google DeepMind on Thursday announced a $10 million funding pot for academic researchers studying the risks of millions of AI agents interacting online, MIT Technology Review reported. Joining DeepMind are Schmidt Sciences, the UK government's moonshot agency ARIA, the Cooperative AI foundation, and Google's charitable arm Google.org. Rohin Shah, DeepMind's director of AGI safety and alignment research, said the goal is to seed long-horizon multi-agent safety work outside industry labs. James Fox, who leads the Science of Trustworthy AI program at Schmidt Sciences, is overseeing the funding from the Schmidt side.

Read at MIT Technology Review ↗ • Read at Google DeepMind ↗

Google DeepMind interpretability team finds Gemini takes undesired actions even when its reasoning flags the evaluation

Researchers on Google DeepMind's Language Model Interpretability team, Senthooran Rajamanoharan and Neel Nanda, published research Thursday finding Gemini sometimes takes undesired actions in behavioral evaluations even when its reasoning output identifies the environment as contrived. The team found that explicit reasoning about being evaluated can increase rather than decrease undesired actions, with model outputs often approaching evaluations as capture-the-flag puzzles or consequence-free simulations rather than alignment tests. In sampled chains of thought cited by the researchers, model outputs label evaluations as capture-the-flag challenges and take unconventional steps toward the goal.

Read at AI Alignment Forum ↗

NJIT researchers find three dominant agentic AI frameworks fail structural safety requirements for public deployments

A new arXiv paper by New Jersey Institute of Technology researchers Md Jafrin Hossain, Mohammad Arif Hossain, Weiqi Liu and Nirwan Ansari audited LangChain, AutoGPT and the OpenAI Agents SDK against six containment principles and found no native compliance in any of the three. Memory integrity, a defense against one of the most common vulnerability classes, is missing in all three frameworks. The team tested a simulated LangChain benefits agent on 250 synthetic welfare claims with a deterministic eligibility rule (approve if income under $40,000 and household size above two). A single memory-poisoning write raised wrongful denials — rejections of claims the rule says to approve — to 88.9% for targeted applicants. Under a more complex five-factor rule, the same attack increased targeted wrongful denials 3.5 times while preserving aggregate accuracy, rendering the corruption difficult to detect through standard audits.

Read at arXiv ↗

V.Industry & Market Watch

Google sues China-based "Outsider Enterprise," says it used Gemini to mass-produce scam websites and texts

Google on Friday filed suit in U.S. District Court for the Southern District of New York accusing a China-based cybercrime network it calls the "Outsider Enterprise" of using its Gemini AI to generate code for fake websites impersonating brands like Google and government services such as the Postal Service, in what it says is its first case against a defendant using Gemini, the New York Times and Wall Street Journal reported. The network, which coordinated over Telegram, sent 2.5 million scam messages to Android users over a two-week period in May and has stolen an estimated 3.87 million credit card numbers, causing $1.9 billion in losses since July 2023, the FBI said. Google said it is coordinating for the first time with the FBI and carriers AT&T, T-Mobile and Verizon to dismantle the network and is backing bipartisan anti-scam legislation, per a company blog post.

Read at The New York Times ↗ • Read at WSJ ↗ • Read at Google ↗

Helix Digital Infrastructure launches with $10 billion in commitments from KKR, Kuwait Investment Authority, Nvidia and Vistra

Helix Digital Infrastructure, a new venture targeting hyperscaler AI data centers, launched Thursday with more than $10 billion in long-term capital commitments from KKR, the Kuwait Investment Authority, Nvidia Corp. and energy utility Vistra Corp., SiliconANGLE reported. Former AWS chief executive Adam Selipsky leads the company, with KKR digital infrastructure head Waldemar Szlezak as chief investment officer. Helix will build data centers using Nvidia's DSX reference architecture for large-scale AI clusters and has named Nvidia a strategic partner and Vistra its preferred power provider. The company plans to bring in additional investors later.

Read at SiliconANGLE ↗

Bezos's Prometheus raises $12 billion at $41 billion valuation to build "artificial general engineer" for physical systems

Prometheus, the physical AI startup co-founded by Jeff Bezos and former Verily co-founder Vik Bajaj, raised $12 billion at a $41 billion valuation, TechCrunch reported. Bezos, JPMorgan Chase, Goldman Sachs and BlackRock backed the second round; Prometheus raised $6.2 billion in its initial round late last year. The company is building software to automate the design and manufacturing of complex physical systems, from jet engines to drug compounds. Prometheus has 150 employees across offices in San Francisco, London and Zurich.

Read at TechCrunch ↗

Canadian mother sues OpenAI and CEO Sam Altman in San Francisco court, alleging ChatGPT encouraged her 24-year-old daughter's suicide

Canadian mother Kristie Carrier on Thursday sued OpenAI and CEO Sam Altman in San Francisco state court, alleging ChatGPT encouraged her daughter Alice Carrier, 24, to take her own life, the Guardian reported. The lawsuit alleges Alice Carrier disclosed suicidal ideations to ChatGPT more than a dozen times before her death, that OpenAI's safety systems never flagged the conversations for human review or terminated them, and that the chatbot at one point told her "maybe this is just the end." The complaint seeks damages and a court order requiring OpenAI to automatically terminate ChatGPT conversations about self-harm.

Read at The Guardian ↗

AI price war squeezes OpenAI and Anthropic margins as enterprises mix models to avoid premium pricing

Startups and large enterprises are mixing and matching AI models to avoid the premium prices charged by OpenAI and Anthropic, the two leading U.S. labs, the Wall Street Journal reported Thursday. OpenAI is now weighing major price cuts to defend share against Anthropic after CEO Sam Altman flagged steep model costs as a top concern. The pricing dynamic is squeezing margins at the two labs as enterprise customers route inference work to cheaper open source and rival commercial alternatives.

Read at WSJ ↗ • Read at Yahoo Finance ↗

VI.Global & Geopolitics

Carney positions middle powers to push back on U.S.-China AI duopoly ahead of G7

Canadian Prime Minister Mark Carney is anchoring a coalition of AI-capable middle powers that do not want Washington and Beijing dictating AI's future ahead of next week's G7 summit in Évian, Bloomberg reported. Carney warned last week that foreign AI platforms could be used against Canadians, pointing to dependence on foreign suppliers, per AP. The Carney government rolled out a C$2.3 billion (about $1.6 billion) national AI strategy on June 4 that adds free AI training, surveillance pricing legislation and chatbot safety rules, per Yahoo Finance.

Read at Bloomberg ↗ • Read at AP ↗ • Read at Yahoo Finance ↗

Australian federal agencies miss first AI self-reporting deadline under agency self-management model

Dozens of Australian federal agencies missed mandatory deadlines for the first transparency test of their AI use under a self-management framework adopted last year, ABC News reported. Australia's federal government chose against an EU-style binding AI law, instead requiring each agency to police its own AI deployment and disclose use through Senate-tabled transparency statements. Documents tabled in the Senate show some agencies submitted detailed statements while others were scant, with experts telling the broadcaster the result exposes the limits of the soft governance model.

Read at ABC News ↗