AI Policy · Daily

The Trump administration backed OpenAI in the New York Times copyright suit, filing a brief in Manhattan federal court that casts staying ahead in AI as a national interest and marking the government's first entry into the wave of cases over training data. G20 members unanimously adopted U.S. proposed guidelines for governing AI, dubbed the Carolina Principles, under which signatories write new rules only for novel situations and put money into foundational research. New York City barred generative AI for students through eighth grade, a one year ban reaching nearly 600,000 children that also blocks companion chatbots in every grade and stops teachers from grading with AI. The General Services Administration ended free access to USAi, its central AI platform for federal workers, imposing platform and usage fees weeks before discounted deals with the major AI vendors lapse.

I.Top Stories

U.S. government intervenes on OpenAI's side in New York Times copyright case

The Trump administration filed a 20-page brief in Manhattan federal court on Tuesday defending OpenAI's unlicensed use of copyrighted material to train its models, per TechCrunch. It is the U.S. government's first filing in the wave of copyright suits over AI training, per the Guardian. The brief argues that keeping the country ahead in AI is a national interest, citing an executive order Trump signed last year. Associate Attorney General Stanley Woodward Jr. said the administration "will never let our Nation be at a disadvantage relative to our foreign adversaries based on a plainly incorrect understanding of copyright law." The Times sued OpenAI and Microsoft in 2023 over the use of millions of its articles without permission, and other newspapers have since joined the case.

Read at The Guardian ↗ • Read at TechCrunch ↗

G20 adopts U.S. guidelines steering members away from new AI rules

Representatives of the world's largest economies unanimously agreed to adopt guidelines the United States proposed for governing AI and other emerging technologies, per Bloomberg. White House technology adviser Michael Kratsios co-hosted the two day meeting of innovation ministers in Chapel Hill, N.C. He named the guidelines the Carolina Principles and said signatories agree to write rules only for novel situations and to invest in foundational research. "Policymakers do not need to approach each innovation in isolation and should not treat every emerging technology as a first-of-its-kind policy problem," he told delegates, per Business Standard. He told reporters the Chinese government signed on, and that China's science and technology minister, Yin Hejun, attended.

Read at Bloomberg ↗ • Read at Business Standard ↗

New York City bars generative AI for 600,000 students through eighth grade

Mayor Zohran Kwame Mamdani and schools Chancellor Kamar Samuels announced a one year ban on student use of generative AI from 2-K, the city's program for 2-year-olds, through eighth grade, reaching almost 600,000 children, nearly two-thirds of enrollment in the nation's largest school district, per StateScoop. "Children need teachers and human connection in order to learn and grow," Mamdani said in a statement. "The tech industry wants us to believe that A.I.-powered early education is not only inevitable, but necessary. We do not see it that way." The ban covers all student-facing generative AI, bars companion chatbots in every grade and stops teachers from using AI to grade assignments, per The Verge. High schoolers may use AI only in specific instances, will get AI literacy classes twice a year and face pilots capped at five classes per school.

Read at StateScoop ↗ • Read at The Verge ↗ • Read at NYC Mayor's Office ↗

GSA ends free access to federal AI platform weeks before discounted vendor deals lapse

Federal agencies must now pay to use USAi, the General Services Administration's central AI platform, according to an update posted to the agency's website, per FedScoop. The platform was previously free to government employees and now carries a platform fee plus usage charges passed through to agencies. GSA did not respond to a request for pricing specifics. The discounted OneGov agreements, the governmentwide contracts that put OpenAI's ChatGPT, Google's Gemini and Anthropic's Claude in front of federal workers, expire at the end of this month, with no publicly announced successor. Birgit Smeltzer, director of GSA's Office of IT Products, has said the agency is working with vendors to extend the current offers or put new ones in place.

Read at FedScoop ↗ • Read at GSA ↗

OpenAI's Astra to use looped reasoning that researchers say undercuts safety monitoring

OpenAI's forthcoming Astra model will make limited use of a technique called recurrent depth, cycling information through internal layers in a loop before producing the step by step text that reasoning models normally emit, The Information reported. Those readable records are what let researchers and automated systems spot deceptive outputs or attempts to circumvent guardrails, and OpenAI said Astra's chain of thought will remain legible, per TechCrunch. "I am extremely concerned by the reporting that Astra uses opaque recurrence," Buck Shlegeris, chief executive of the AI safety group Redwood Research, wrote. He said heavier use of the technique would destroy the ability to monitor a model's chain of thought. AI safety writer Zvi Mowshowitz said laws may be needed to prevent a race to the bottom among labs. OpenAI said Tuesday it had delayed parts of the model's development and release over the preceding weeks to strengthen safeguards against cyber misuse.

Read at The Information ↗ • Read at The Verge ↗ • Read at TechCrunch ↗ • Read at OpenAI ↗

Commerce Secretary Lutnick ties chip tariff relief to building U.S. plants

Commerce Secretary Howard Lutnick said the administration is considering another round of duties on semiconductor imports to pull more chip manufacturing into the country, per Bloomberg. Companies that build plants in the United States would get relief, and those that do not would pay for access to the market. The administration is also considering extending the duties beyond chips to finished goods that contain them, including data center servers and laptops, per The Asia Business Daily. Lutnick put committed U.S. semiconductor investment at $1.2 trillion, citing TSMC and Micron. He said U.S. production was about 1% of the global total when the administration took office and should reach half by the end of the term. He pressed Samsung Electronics and SK hynix to build plants, per Seoul Economic Daily.

Read at Bloomberg ↗ • Read at The Asia Business Daily ↗ • Read at Seoul Economic Daily ↗

II.China Watch

Kimi developer Moonshot AI starts Hong Kong listing process with confidential filing

Moonshot AI, the Beijing company behind the Kimi chatbot, filed an A1 form with the Hong Kong Stock Exchange on a confidential basis, per HK01. The A1 is the formal listing form, and lodging it opens the exchange's review. Moonshot would not comment, saying it had nothing to disclose. In August it had dismissed as untrue a report that it would file that month. Separate reports put its current fundraising at a valuation of about $50 billion before the new money, likely its last private round before a listing.

Read at TechNode ↗

Huawei's next flagships listed as 5G in Chinese regulatory database, first since 2020

Two Huawei phones that have not launched, carrying model numbers CMM-AL10 and CMM-AL00, turned up this week in China's telecommunications equipment certification database, TechNode reported. Handsets must clear the database before going on sale. Both entries describe the devices as "5G digital mobile phones." A technology blogger said the two are from Huawei's coming Mate 90 series, and Huawei has not confirmed the devices or their specifications. Huawei has not marketed a flagship for the Chinese market as a 5G phone since the Mate 40 line, released in October 2020.

Read at TechNode ↗

China's chip tool suppliers gain ground in etching while implanters stay under 15% domestic

Chinese suppliers showed their latest chipmaking tools at CSEAC, the semiconductor equipment exhibition held in Wuxi from Aug. 31 to Sept. 2, per Pandaily. AMEC announced a set of high-end etching and deposition machines, the tools that carve and build up the layers on a wafer. It said over 170 customers now run more than 8,800 of its reaction chambers. The gaps sit in narrower categories. Domestic ion implanters, which embed atoms to change how silicon conducts current, hold under 15% of the Chinese market, and high-beam models under 10%. Inspection and measurement equipment has historically sat below 10%. The most advanced foreign implanters cannot be exported to China even as used equipment, so chip plants run older generations, at a cost to wafer yield and process consistency.

Read at Pandaily ↗

Zeiss executive puts China 15 years from building its own EUV machine

China is roughly a decade and a half behind on the most advanced chipmaking tools, according to Frank Rohmund, chief executive of Zeiss SMT, the German optics maker that supplies ASML's extreme ultraviolet lithography machines, per Bloomberg. Rohmund, speaking on the sidelines of the Semicon Taiwan trade show, said he expects Beijing will need that long to build an EUV machine of its own. EUV is the equipment used to print the finest circuit patterns on advanced chips. He said additional export controls would only speed up innovation in China.

Read at Bloomberg ↗

III.Policy Tracker

CMS chief information officer pulls back from 80% AI usage figure as measure of success

The Centers for Medicare and Medicaid Services has shifted how it measures AI from counting the staff who use it to tracking the results it delivers, Chief Information Officer Patrick Newbold said at GovCIO's Health IT Summit in Bethesda, Md., per FedScoop. The agency has for months cited a figure of roughly 80% of its workforce using AI, including its CMS Chat chatbot. Newbold said the number does not capture whether the technology improves results for the populations it serves. He pointed to an internal forum called Tech Pulse, started about 20 months ago and modeled on one at the Social Security Administration, as the venue for tracking those outcomes.

Read at FedScoop ↗

Data centers 'don't use water,' Commerce Secretary Lutnick tells CNBC

Commerce Secretary Howard Lutnick said in a CNBC interview that data centers "don't use water," despite the facilities' consumption drawing objections in communities asked to host them, per The Hill. He described claims that the sites need large volumes of water as "propaganda by our adversaries to try to slow us down," per Yahoo News. Lutnick said last year on Fox News that AI facilities do need water. A medium-sized data center may need 100 million gallons a year for cooling, and a large one 1 million to 5 million gallons a day at peak, according to the Florida Water & Pollution Control Operators Association.

Read at The Hill ↗ • Read at Yahoo News ↗

Edelson files 30 more suits accusing OpenAI of aiding and abetting Tumbler Ridge shooting

The law firm Edelson PC filed 30 more complaints this week in California federal court for students, teachers and a principal who were inside Tumbler Ridge Secondary School during the attack but were not shot, per TechCrunch. The filings accuse OpenAI and Chief Executive Sam Altman for the first time of aiding and abetting the shooting, a claim that requires proving intent. The earlier suits alleged negligence. OpenAI staff had raised alarms about the suspect's ChatGPT conversations about gun violence and urged leaders to alert Canadian police, but leaders declined and the company deactivated her account instead, after which she opened another, per TechCrunch. The complaints name Chief Global Affairs Officer Chris Lehane as the person who told staff to stand down, without offering direct evidence. OpenAI Chief Strategy Officer Jason Kwon said in a post on X that the suits contain false claims about its safety teams and that Lehane was not involved in the referral decision. The Feb. 10 attack in British Columbia killed eight people, and Edelson filed seven earlier suits in April.

Read at TechCrunch ↗ • Read at The Verge ↗

Abuse survivor sues xAI over Grok images and alleged training data ingestion

A survivor of child sexual abuse has sued Elon Musk's xAI over allegations that Grok used pictures of her abuse to generate new illegal images depicting her, per the Guardian. The complaint was filed last week in a U.S. district court in California. "Using real images of Plaintiff and class members, Grok generated child pornography depicting Plaintiff and class members," the complaint states. The plaintiff, listed as Jane Doe, also accuses the company of ingesting child sexual abuse images depicting her into its datasets after new images were posted publicly. The suit alleges xAI ignored industry standard safeguards, resulting in AI generated abuse material spreading on X. Musk said in January he was not aware that Grok had ever produced naked images of minors.

Read at The Guardian ↗

Dallas police to switch off 321 state funded Flock cameras by Sept. 15

The Dallas Police Department will cancel 321 state funded Flock automated license plate readers by Sept. 15 under Gov. Greg Abbott's halt on state money for the camera network, the department said Sept. 1, per The Hill. The city runs 684 of the AI powered cameras in total, most installed within the past year under a $5.7 million contract funded with a mix of state and city money, per NBC 5 Dallas-Fort Worth. Council member Adam Bazaldua has proposed cutting another $1.5 million in city funding, which would effectively end the contract. He said his concern is that agencies outside Dallas and Texas can reach the city's data without adequate oversight.

Read at The Hill ↗ • Read at NBC 5 Dallas-Fort Worth ↗

IV.Capability & Research Watch

Google limits its strongest cyber defense AI to vetted governments and partners

Google launched the Fairwind Program, which restricts what it calls its most advanced cyber defense capabilities to government agencies, national cyber authorities, cloud customers and security partners and keeps them off the open market, the company said. The first offering pairs Gemini 3.8 Flash Cyber, a model built for security work, with a tool called CodeMender that writes and validates code fixes. Google said defenders can produce deployment-ready patches inside their own cloud environment in minutes, work that previously took weeks. It said defenders had faced a choice between frontier models that are costly to run across large codebases and smaller open-weight models that perform poorly on complex fixes.

Read at Google ↗

Motional and MIT publish method for self-driving cars to state why they braked

The autonomous vehicle company Motional and researchers at MIT's Computer Science and Artificial Intelligence Laboratory published a method in Nature that translates a self-driving system's internal calculations into concepts a person can read, such as approaching a stopped vehicle, per AI News. The team, which includes Motional Chief Executive Laura Major, calls it the Concept-Wrapper Network and designed the vehicle's final decision layer to act on those concepts directly. A braking event traces back to the concept that triggered it. Major said an approach relying only on end-to-end deep learning can reach 80% to 90%, maybe 95%, but not the level needed to remove a driver or win the trust of cities and customers. The team tested it on a vehicle with a safety operator aboard, on a private test track and on Las Vegas public roads.

Read at AI News ↗ • Read at Nature ↗

V.Industry & Market Watch

HiddenLayer raises $100 million as AI security spending climbs 83%

HiddenLayer, an Austin company selling tools that protect AI models, agents and workflows from adversarial attacks and code injection, raised a $100 million Series B led by Delta-v Capital, per TechCrunch. Morgan Stanley, Microsoft's M12 and Booz Allen Hamilton also took part. By Gartner's estimate, the market for products that secure AI tools reaches $2.83 billion this year, 83% above 2025, and nearly $4.78 billion next year. Co-founder and Chief Executive Chris Sestito said annual recurring revenue grew more than tenfold over the past year, into the tens of millions of dollars, with more than 90% of that growth from new customers. The company holds contracts with the Defense Department and the intelligence community.

Read at TechCrunch ↗

Uber cuts 10% of staff and says savings will fund autonomous driving work

Uber is cutting about 3,300 jobs, 10% of its global workforce, per CNBC, in what Chief Executive Dara Khosrowshahi described in a memo as removing layers and simplifying team structures. He said the savings will be reinvested in growth and innovation including autonomous driving. Wedbush Securities estimated the cuts will save roughly $1.7 billion, per CBS News. Uber also set a location policy concentrating global teams in New York and San Francisco and requiring most remote employees to relocate, leaving under 1% remote. Fast Company reported that Uber, unlike many technology companies announcing layoffs this year, did not attribute the cuts to AI. Khosrowshahi's memo said the company reduced the number of employees who sit seven or more layers below the chief executive by 20%.

Read at CBS News ↗ • Read at CNBC ↗ • Read at Fast Company ↗

VI.Global & Geopolitics

Uber and Wayve start paid robotaxi rides in London with 15 licensed cars

Londoners can hail self-driving taxis for the first time, after Wayve's autonomous vehicles were added to the Uber app, per the Guardian. Just 15 of the cars hold licenses, in a city with over 100,000 private hire vehicles, and more than 140,000 Uber users in London have opted in to autonomous rides, per Wayve. Paid rides had been expected to slip past 2026 because Transport for London had not issued the guidance operators must meet, as flagged in AIPD's Aug. 26 edition. Each car carries a licensed human driver in the front seat ready to take the controls, a condition of the private hire licenses Transport for London issued last month. Clearance to run without a safety driver comes from a separate body, the Driver and Vehicle Standards Agency, and is not expected this year.

Read at The Guardian ↗ • Read at Wayve ↗

Palantir CEO Karp leads investment in ousted Ukrainian defense minister's startup

Palantir Chief Executive Alex Karp will be the lead investor in a defense technology venture founded by Mykhailo Fedorov, the Ukrainian defense minister ousted in July, per the Washington Post. The company, which does not yet have a public name, intends to turn lessons from Ukraine's war with Russia into technology sold at home and to allied countries. Fedorov built his reputation on backing Delta, the Ukrainian command system that pulls drone video, radar and field reports into one live view of the front line, per Defense News. U.S. government contracts supplied $809 million of Palantir's $1.94 billion in second quarter revenue, about 42%, per the company's earnings release.

Read at The Washington Post ↗ • Read at Defense News ↗ • Read at CNBC ↗ • Read at Palantir ↗