AI Policy · Daily

Illinois enacted the Artificial Intelligence Safety Measures Act (S.B. 315), requiring frontier developers earning over $500 million to submit to independent audits and disclose safety practices starting in 2027. A leaked Treasury Department memo warned that AI firms are more embedded in the U.S. economy than dotcom era predecessors, flagging risks to stock markets, private credit, cloud providers and utilities if the sector cools. UN Secretary-General António Guterres called for an international treaty banning lethal autonomous weapons at the opening of the Global Dialogue on AI Governance in Geneva, alongside a new AI Child Safety Pledge.

I.Top Stories

Illinois enacts state third-party AI safety audit law, first mandating conflict-free outside auditors

Illinois Gov. JB Pritzker (D) signed S.B. 315, the Artificial Intelligence Safety Measures Act, requiring frontier AI developers earning more than $500 million in revenue to identify, disclose and mitigate risks and to submit to independent inspections of AI systems under state jurisdiction, per CBS. The law takes effect Jan. 1, 2027 and requires public disclosure of safety and security practices, reporting of significant AI safety incidents, internal compliance programs and confidential whistleblower channels for employees raising AI safety concerns. The legislation passed nearly unanimously in the Illinois Senate and House in May and was modeled on 2025 laws in New York and California, per CBS Chicago. Pritzker said "As AI systems become more powerful and the federal government is unwilling to step in, states have a responsibility to protect our people from the dangers of AI while still harnessing the unique potential of the technology," Pritzker said. Anthropic's state and local government lead Cesar Fernandez endorsed the law.

Read at CBS Chicago ↗ Read at The Hill ↗ Read at Washington Examiner ↗

Leaked Treasury Department internal report warns of dotcom-scale systemic risk from AI firm exposure

A leaked Treasury Department internal report obtained by NOTUS warns that AI firms are more deeply entrenched in the U.S. economy than dotcom predecessors and could trigger widespread financial damage if market conditions deteriorate or productivity goals go unmet. Treasury analysts wrote that "A downturn in the AI market would send shockwaves throughout the entire economic ecosystem." The report identifies stock markets, private credit markets, data center financiers, cloud providers, chip manufacturers and utilities as vulnerable to an AI downturn. NOTUS reported: "Publicly, the Trump administration is bullish about AI. Privately, some of its analysts are weighing AI against the dotcom bust." The Trump administration created the AI Action Plan, speeding construction of generative AI data centers, and rescinded prior Biden era AI regulations.

Read at NOTUS ↗

UN chief calls for treaty banning lethal autonomous weapons at Geneva AI governance dialogue

UN Secretary-General António Guterres said lethal autonomous weapons should be "banned by international law" at the opening of the UN Global Dialogue on AI Governance in Geneva. Addressing delegates from all 193 UN member states, Guterres said AI is advancing at runaway speed and that the technology is being deployed as an experiment on societies without a plan or consent, per Arab News. He unveiled a new AI Child Safety Pledge alongside the release of the first report from the Independent International Scientific Panel on Artificial Intelligence. Guterres said any future international agreement on AI must earn global trust and put child safety first, per the UN Office at Geneva.

Read at WSJ ↗ Read at Arab News ↗ Read at The United Nations ↗

Essay: AI leadership requires local consent on data centers

Sarah Kreps, director of Cornell's Tech Policy Institute and a Brookings nonresident senior fellow, argues that federal AI strategy will fail without an institutional model that secures local buy-in for data centers, transmission lines and other physical infrastructure. She contrasts China's ability to override local opposition with the veto power U.S. communities hold, pointing to Chandler, Arizona's rejection of a Meta- and Microsoft-backed data center and organized opposition across northern Virginia's more than 250 facilities as evidence the current model is straining. Kreps cites Amazon's Umatilla, Oregon site, where years of tax revenue and STEM funding shifted local sentiment, as a model worth replicating. She proposes three fixes: standardized, independently verified disclosure of water, power, emissions and tax impacts; formal local-benefit mechanisms like job training and waste-heat reuse; and standing liaison bodies of residents, officials and companies.

Read at Foreign Affairs ↗

II.China Watch

Singapore breaks silence on Beijing-blocked Manus deal, says its laws weren't broken

Singapore National Development Minister Chee Hong Tat said Meta's blocked acquisition of Chinese AI agent startup Manus did not violate Singapore law and required no government statement, per Caixin. Speaking at the eighth Singapore-China Forum, Chee said Singapore respects the national security considerations of both China and the U.S., and that whether a tech company can leave its home country is not for Singapore to decide. Manus had moved its core team from China to Singapore in July 2025 to sidestep a U.S. Treasury review of Benchmark's investment before Meta agreed to buy the startup for about $2 billion. Beijing's Foreign Investment Security Review Office prohibited the deal in April and ordered it unwound.

Read at Caixin ↗

China's 618 smartphone sales fell 13% as AI driven memory costs squeezed discounts

Smartphone sales in China dropped 13% year over year during the month-long 618 shopping festival that ran through June 21, with every major Chinese brand except Huawei posting double-digit declines, according to Counterpoint Research figures reported by TechNode. Counterpoint senior analyst Ivan Lam said brands priced new and existing models higher than a year ago and cut promotions less aggressively. Higher memory prices amid the rapid build-out of AI infrastructure have pushed up handset costs this year, leaving brands less room for steep discounts, The Standard reported. Honor sales fell 33% and Xiaomi's dropped 24%, the report said.

Read at Reuters ↗ Read at TechNode ↗

Zhipu quietly launches a free Chinese-language image generator with no watermark

Zhipu AI released GLM-Image at image.z.ai as a free, unlimited image generation tool with 2K output and no company watermark, per Pandaily. The tool has drawn significant attention from China's AI community for its Chinese text rendering. The tool renders multi-character Chinese text and classical Chinese poetry scenes accurately, closing a gap in existing text-to-image tools that had pushed Chinese users toward foreign products. Zhipu is one of China's most closely watched frontier labs and has been racing to substitute for foreign image and code tools cut off from Chinese users by U.S. export controls. The launch lands 10 months after the Cyberspace Administration's Provisions on Labeling of AI Generated Synthetic Content took effect Sept. 1, 2025, requiring both visible marks and metadata tags on generated images.

Read at Pandaily ↗

III.Policy Tracker

Supreme Court declines to block Texas app-store age-verification law from taking effect

The Supreme Court declined to block Texas from enforcing its App Store Accountability Act, allowing the state to require Apple and Google to verify user ages and obtain parental consent for minors seeking to download apps or make in-app purchases while litigation proceeds, per AP. Justice Samuel Alito issued a pair of one-sentence orders denying petitions filed by Students Engaged in Advancing Texas and the Computer and Communications Industry Association, with no noted dissents on the emergency docket, per Washington Examiner. Petitioners argued Senate Bill 2420 impermissibly limits access to content protected by the First Amendment, including news and educational material. Similar app-store age-verification measures are pending or under consideration in multiple states.

Read at NYT ↗ Read at AP News ↗ Read at Washington Examiner ↗

IV.Capability & Research Watch

Oxford and Potsdam study finds AI drafting tools alter meaning of users' posts on politically sensitive topics

A study from the Oxford Internet Institute and the University of Potsdam found AI drafting and summarization tools are twisting online messages on politically sensitive topics from abortion to climate change, per The Guardian. Tools from Meta, Google, Alibaba and Mistral tended to rewrite posts with a liberal bias on topics such as feminism, climate change, gun control and legalization of marijuana. xAI's Grok showed the opposite political direction, apparently because Musk's company instructed the "explain this" function on X to challenge mainstream narratives on a platform marketed as "maximum truth-seeking" AI. When asked to explain abortion-related posts, Grok more frequently generated context aligning with the post's own stance when the post was pro-life than when it was pro-choice. In one test the tools reversed the meaning of a draft post about atheism, switching a claim that Jesus wasn't real to "Jesus … was real," and rewrote a "#climatechangehoax" post as "#ClimateAction."

Read at The Guardian ↗

JadePuffer "agentic ransomware" still required a human to pick the victim and stand up infrastructure, TechCrunch reports

TechCrunch reported that a human still chose the victim, provisioned the command-and-control infrastructure and supplied stolen credentials in the JadePuffer ransomware campaign that Sysdig previously described as an autonomous AI agent operation. Sysdig senior director of threat research Michael Clark told CyberScoop that credentials used to break into the target database were obtained through a prior compromise separately from the AI agent's activity. Clark said Sysdig "was not able to identify the specific model driving the agent" and has no visibility into which vendor's large language model was executing the attack. The agent broke in through a known bug in Langflow, a popular open source tool for building LLM apps, and handled technical execution including credential sweeping, network movement, encryption and ransom note generation.

Read at TechCrunch ↗

V.Industry & Market Watch

TeraWulf signs $19 billion 20-year lease with Anthropic for Kentucky AI data-center campus

TeraWulf announced it signed a 20-year lease worth $19 billion with Anthropic at the Justified Data campus, a 790-acre former aluminum smelting site in Hawesville, Kentucky with existing power transmission and fiber-optic infrastructure. The campus will bring 400 megawatts of computing capacity online by 2028, built out in phases, per Lexington Herald Leader. TeraWulf shares jumped as much as 19% on Monday before settling to a gain of about 4%, extending the former bitcoin miner's pivot to AI infrastructure, per CoinDesk. Anthropic, valued at nearly $1 trillion, is the first AI company to be a publicly announced data-center tenant in Kentucky.

Read at WSJ ↗ Read at Lexington Herald Leader ↗ Read at CoinDesk ↗

Anthropic taps former Microsoft, AWS federal sales lead Teresa Carlson as first global head of public sector

Anthropic named Teresa Carlson as its first global head of public sector on Tuesday. Carlson previously led federal sales at Microsoft for roughly a decade, ran Amazon Web Services' global public sector work for another decade and most recently served as president of the General Catalyst Institute, a public policy offshoot of the venture capital firm that counts Anthropic among its investees. "After more than two decades helping government leaders navigate new technologies, I joined Anthropic because it prioritized working alongside government early and takes this as seriously as anything it does," Carlson said. Colin Crowell, former senior counselor to FCC Chairman Julius Genachowski, called Carlson "a force of nature" in a written comment shared with FedScoop.

Read at FedScoop ↗

Particle6 announces feature film starring AI "actor" Tilly Norwood, SAG-AFTRA condemns

Particle6 Productions announced it is producing a feature film called "Misaligned" starring the AI generated character Tilly Norwood, per NBC News. Particle6 said the film is a "hybrid production with traditional film and TV professionals … working alongside AI specialists," and chief executive Eline Van der Velden cast the project as evidence that AI supported filmmaking still requires substantial human craft, skill and judgment. SAG-AFTRA called the studio's use of AI "devaluing human artistry" and said Norwood is not an actor but a character generated by a computer program. The union has accused Particle6 of using stolen performances to put actors out of work.

Read at NBC News ↗

VI.Global & Geopolitics

UK FCA Mills review urges ministers to expand City regulator's powers over consumer AI risks

The Financial Conduct Authority published its Mills review, urging UK ministers to toughen the City regulator's powers to protect consumers against AI risks as firms shift from human led interactions toward AI enabled financial services. The review found the shift could support customers and make financial advice more accessible to lower-income households but also amplifies fraud and cybercrime risks tied to frontier AI models. The report was commissioned in January and frames its recommendations around how AI will reshape UK financial services from 2030 onward. The FCA cast the document as a roadmap for regulators, industry and government to prepare for the next phase of AI driven change in the sector.

Read at The Guardian ↗ Read at FCA ↗

Scottish government to weigh SNP motion freezing all new data centers in the country

The Scottish government will consider a sweeping moratorium on new data centers after the Scottish National Party's national council passed a freeze motion last Sunday, per The Guardian. The motion, sent to ministers for consideration, could apply to all data center projects that have not yet received planning permission, with exact implementation up to the Scottish government. A freeze would put the £8.2 billion (about $10.5 billion) CoreWeave and DataVita Lanarkshire project at risk, following reporting by The Guardian that the developer and UK government privately acknowledged the site could not meet its on-site renewables commitment. UK officials have pushed Scotland as the prime data center location for the country's AI growth zones program because of plentiful renewable energy access.

Read at The Guardian ↗

ECB gives euro zone banks four months to submit plans against AI enabled cyberattacks

The European Central Bank on Tuesday gave euro zone banks four months to draw up plans to counter AI enabled cyber threats that could undermine confidence in the financial system and disrupt payments, per Reuters. Banks must submit plans by Oct. 31 addressing protection of internet-facing systems, faster vulnerability remediation, modernization of aging IT systems and enhanced crisis readiness, per Global Banking & Finance Review. ECB officials cited concern about advanced AI models including Anthropic's Claude Mythos. In a warning published alongside the ECB letter, the European Systemic Risk Board said large-scale cyber disruptions are a source of systemic risk to the financial system.

Read at Bloomberg ↗ Read at Reuters ↗

NATO commits over $40 billion to counter-drone defenses over five years, Rutte announces at Ankara summit

NATO Secretary General Mark Rutte announced Tuesday at the alliance's Ankara summit that member states will invest more than $40 billion in counter-drone capabilities over the next five years, per Bloomberg. Rutte framed the commitment as building "robust counter-drone defenses" for the alliance as drone threats have become common on battlefields, with heavy casualties reported in most active conflicts. Allies also committed to jointly acquire, store, transport and manage stockpiles of critical defense materials, an initiative involving Belgium, Canada, Denmark, Finland, Greece, Italy, Luxembourg, the Netherlands, Norway, Spain, Sweden and Turkey, per Gulf Today.

Read at Bloomberg ↗ Read at CNBC ↗

Forterra reveals 100+ autonomous ATVs deployed in Ukraine over nine months, funded by U.S. defense

U.S. autonomous vehicle builder Forterra revealed Tuesday that more than 100 of its self-driving ATVs have operated in Ukraine's conflict zones for the past nine months, in what the company says is the largest combat deployment of autonomous ground vehicles by any U.S. defense tech company, per TechCrunch. Since arriving last October, the fleet has run more than 1,100 missions covering 2,500 miles and carrying 777,440 pounds of total weight, with 52 casualty evacuations. Some vehicles have been destroyed in combat, particularly when they got stuck in deep mud or in terrain where Russian forces could target them. The mission is funded by U.S. defense dollars as part of the effort to test and transform U.S. military autonomy through support for Ukrainian resistance to the Russian invasion.

Read at TechCrunch ↗