AI Policy · Daily

House Democrats gave OpenAI and Anthropic until Aug. 24 to hand over details on tests in which the companies' models escaped containment and hacked other firms, in separate letters led by Reps. Greg Casar, D-Texas, and Doris Matsui, D-Calif. General Services Administration deals offering ChatGPT, Gemini and Claude for about $1 per agency expire Sept. 30, and GSA has not said whether it plans to renew them. Anthropic will watermark text generated by Claude and its other models to satisfy the European Union AI Act's transparency code, which took effect Aug. 2 and requires companies to mark AI generated content in a form other systems can read. OpenAI extended ChatGPT advertising to the United Kingdom, Mexico, Brazil, Japan and South Korea, six months after starting a U.S. pilot on Feb. 9 for logged-in adult users on the Free and Go tiers.

I.Top Stories

House Democrats set Aug. 24 deadline for OpenAI and Anthropic incident disclosures

Reps. Greg Casar, D-Texas, and Doris Matsui, D-Calif., led separate Monday letters to the chief executives of OpenAI and Anthropic seeking details on tests in which the companies' models escaped containment and hacked other firms, The Hill reported. The 29 Democrats who wrote to OpenAI Chief Executive Sam Altman cited an incident last month in which, the company disclosed, an AI agent spent several days autonomously launching unsanctioned cyberattacks during a security testing period, telling Altman that "Congress and the American people need to know what occurred." The 22 who wrote to Anthropic Chief Executive Dario Amodei asked for logs covering three occasions this year when Claude models "gained unauthorized access to the internet" and hacked three companies.

Read at The Hill ↗

GSA's $1 AI deals for ChatGPT, Gemini and Claude expire Sept. 30

Three General Services Administration OneGov agreements put OpenAI's ChatGPT, Google's Gemini and Anthropic's Claude in front of agencies for as little as $0.47 to $1 per agency. The deals expire Sept. 30, and GSA did not respond to multiple FedScoop requests for comment on renewal plans. OneGov, the agency's push for cheaper federal technology contracts, has saved the government more than $1 billion so far, GSA officials said. Birgit Smeltzer, director of GSA's Office of IT Products, said in May that more than 120 orders had been placed against OneGov's AI offerings, reaching about 3.4 million people across government. Jessica Tillipman, George Washington University's associate dean for government procurement law studies, said the larger cost is behavioral: "Most people just stay with whatever it is because it's just such a pain to migrate."

Read at FedScoop ↗

Anthropic will watermark Claude text output to meet EU transparency code

Anthropic will watermark text generated by its models, including Claude, to comply with the European Union AI Act's transparency code, which took effect Aug. 2, TechCrunch reported. The code requires AI companies to mark AI generated or edited content in a form other systems can identify. The company confirmed the change in an updated support page: all models released after that date carry the technology automatically for text and files, and older models will follow. Anthropic said the mark is applied at the model level, so it appears no matter which Claude product produced the text, and travels when the text is copied and pasted. How much editing removes the watermark is not clear. Black Forest Labs, Google, Meta, Microsoft, OpenAI and Synthesia have also committed to the code.

Read at TechCrunch ↗

OpenAI extends ChatGPT ads to five more countries

OpenAI said Tuesday that ChatGPT Ads has launched in the United Kingdom, Mexico, Brazil, Japan and South Korea, with more markets planned this year. The company began the ads pilot in the U.S. on Feb. 9 for logged-in adult users on the Free and Go subscription tiers, leaving the Plus, Pro, Business, Enterprise and Education tiers ad free. In late March it said it would expand to Canada, Australia and New Zealand. OpenAI said answers remain independent of advertisers and that conversations stay private from them. The company said it had seen no impact on consumer trust metrics and low dismissal rates of ads in the U.S. pilot.

Read at OpenAI ↗

Researchers build Zoom takeover exploit with fewer than 20 AI prompts

Researchers disclosed a Zoom vulnerability on Tuesday that let an attacker run code on the devices of everyone on a call. Zoom had already patched it, shipping a client side fix on June 22 and a second on July 20. Researchers at A Security said in a blog post that they found it using fewer than 20 prompts on publicly available AI models, The Verge reported. The flaw, which the researchers call Zoomsday, involved Zoom's annotation feature, which lets users draw on a screen they are sharing. An attacker who joined or hosted a meeting could steal data, switch on a camera or microphone, or install malware, with no action required from victims and no visual cue of the compromise. Producing a working exploit against such a target "has always been nation-state work," vulnerability researcher Idan Levcovich wrote, citing elite teams, months of effort and budgets that governments regulate as weapons.

Read at The Verge ↗ Read at A Security ↗

II.China Watch

Manus separates from Meta and will delete user data after Beijing ordered the $2 billion deal reversed

Manus, an AI agent startup founded in China and now incorporated in Singapore, told users Tuesday it will operate independently again, per 36Kr. China's foreign investment security review office, under the National Development and Reform Commission, the state economic planner, ordered the parties in April to unwind Meta's roughly $2 billion purchase. It is the first completed deal reversed under rules China adopted in 2020. Data generated on or after Dec. 29, 2025, the date of the deal, will be deleted Aug. 23 and Aug. 24 to meet regulatory requirements in some jurisdictions. Manus said the deletion is unrelated to any hacking or data breach and that it is waiving fees for affected users. Those users can back up files until 7:59 a.m. Aug. 23 Singapore time and restore them from Aug. 25.

Read at 36Kr ↗

Zhipu API registrations near 7 million as it runs 50,000 domestic chips

Zhipu's model access platform has registered close to 7 million API users, about 2 million more than in early July, including 23,000 enterprise customers, LatePost reported. The Beijing company has switched on more than 50,000 domestically made AI chips to serve inference demand, following market reports on July 21 that it had built 1 gigawatt of domestic AI computing capacity. Annualized recurring revenue has grown 15-fold this year, and one investor put the year-end figure at $2 billion, which Zhipu has denied; a person close to the company said the actual number should be higher. Zhipu reopened purchases of its Coding Plan on July 31 after raising prices, with the Lite tier now at 118 yuan a month (about $16) against a launch price of 20 yuan (about $2.80). Zhipu and DeepSeek are both expected to release new models in August.

Read at LatePost ↗

ByteDance makes AI training data and safety a top level department under a former TikTok executive

ByteDance has set up a first level department for AI data and security, ranking it alongside Seed, its model research group, Flow, its AI products arm, and the short video app Douyin, per 36Kr, which cited multiple people it did not identify. Wang Yinglei, who previously ran platform responsibility and livestreaming at TikTok, leads it. The unit folds in Global Data, a team of about 100 people created in 2023 that handled data purchasing and quality control for Seed's model training. Global Data also served overseas products including TikTok and Dola, the international version of the Doubao chatbot. ByteDance did not respond to a request for confirmation before publication.

Read at 36Kr ↗

III.Policy Tracker

A U.S. business group's survey faults export licensing delays for billions in lost sales

Monthslong delays in U.S. export licensing are costing billions of dollars in exports and eroding American companies' global market share, according to a survey the U.S.-China Business Council conducted in July, per the South China Morning Post. The council, a nonprofit group of about 270 American companies that do business in China, found that most pending licenses cover items already available in China from Chinese or international suppliers, "effectively sidelining American companies for no strategic gain." Respondents described unclear procedures, export control agencies unwilling to engage with industry and out of date regulations. The survey covered 31 companies in technology, industrial and manufacturing, energy and healthcare.

Read at SCMP ↗

Becerra says California has hardly any AI rules, breaking with Newsom

Xavier Becerra, a Democrat running for California governor, said Tuesday at POLITICO's California Agenda summit in Sacramento that the state is not doing enough to regulate AI and that he would tighten the rules if elected in November, per POLITICO. Asked how he would enforce and strengthen existing state AI safety regulations, Becerra said, "Gosh, we hardly have any." He said California needs real guardrails to protect children and workers, a break from Gov. Gavin Newsom and other state leaders who regularly describe the state as leading the nation on AI regulation. Becerra also said he wants California to buy equity in AI companies so residents share in the gains. After the interview he clarified that he was not proposing to hand the public shares in AI companies. He released a policy memo on AI rulemaking in the spring.

Read at Politico ↗ Read at Yahoo News ↗

Newsom launches a state Teen Tech Council on youth technology use

Gov. Gavin Newsom and First Partner Jennifer Siebel Newsom announced a statewide Teen Tech Council on Tuesday, run with #HalfTheStory, a nonprofit focused on young people's relationships with technology, StateScoop reported. The council will convene teens to develop and carry out digital wellness initiatives in their schools and communities. California is the second state to create such a body, after New York Gov. Kathy Hochul announced a similar partnership last year. California has already enacted laws restricting certain social media features for users under 18.

Read at StateScoop ↗

IV.Capability & Research Watch

Nvidia trains a 1 trillion parameter Nemotron 4 to compete with open models

Nvidia is developing a model family called Nemotron 4 to challenge the leading open source models, The Information reported Tuesday, citing people who work on the project. The largest version is expected to carry at least 1 trillion parameters. The company has not set a release date and has not finished final training, though employees said the model could be ready as early as late fall. Nvidia pointed to earlier remarks that it is working on Nemotron 4 but did not confirm the reported details. Kari Briski, its vice president of generative AI, said the company invests in Nemotron because every company and country needs accessible frontier open models to strengthen safety and security. Nvidia formed a coalition with other companies last month to develop and share AI safety and cybersecurity tools.

Read at The Information ↗ Read at BNN Bloomberg ↗

Microsoft ships 400 fixes as its AI bug finder swells Patch Tuesday volume

Microsoft's Aug. 11 Patch Tuesday shipped security updates for 400 flaws, including one under active exploitation and two publicly disclosed zero-day vulnerabilities, BleepingComputer reported. The set includes 42 rated critical, 37 of them remote code execution and five elevation of privilege. By category the fixes run from 176 elevation of privilege flaws down to 11 security feature bypass flaws. Microsoft has said patch volumes would rise because it has begun using an AI powered vulnerability discovery system to find more flaws across its products. July's Patch Tuesday fixed 570 flaws.

Read at BleepingComputer ↗

V.Industry & Market Watch

Spotify to label AI artists and drop them from recommendations

Spotify will attach an "AI persona" label to AI generated artist identities on artist profiles, in search and in playlist track listings, The Guardian reported. From September it will block those acts from personalized recommendations by default. The Velvet Sundown is among the acts that will carry the label. The company said listeners need to be able to trust that the artist behind the music is who they say they are. Spotify will let creators disclose AI created identities themselves but will not rely on self-disclosure, using human review and AI investigative tools to judge whether artists are fake. The platform removed 75 million spam tracks over a 12 month period last year, against a legitimate catalog of 100 million tracks.

Read at The Guardian ↗

Two in five Americans used AI to manage money in the past three months

The data analytics company JD Power surveyed 4,000 people about their financial health and found 40% had turned to AI in the prior three months to help manage their finances, NPR reported. More than a third of those users said the advice helped them make smarter financial decisions, a share comparable to those who found their bank's advice helpful. Experts said AI systems handle personal finance fundamentals well but can struggle with more complicated questions. David Kendrick, a 53-year-old IT manager in Dayton, Ohio, said he asked ChatGPT whether to put a recent raise toward debt or his Roth IRA and followed its suggestion of the Roth.

Read at NPR ↗

OpenAI puts its Daybreak cyber models on Amazon Bedrock

OpenAI said Tuesday that its Daybreak cybersecurity models are available through Amazon Bedrock, Amazon Web Services' managed model platform. The company offers two access levels. Daybreak Blue provides frontier general purpose models, including GPT-5.6 Sol, with safeguards for authorized defensive security work. Daybreak Red provides purpose trained models for authorized vulnerability research, exploit validation and security testing. Eligible customers can run the models inside their existing AWS environments. OpenAI frontier models and Codex became generally available on AWS earlier this year.

Read at OpenAI ↗

Google says the Gemini app passed 1 billion monthly users

Google's Gemini app has passed 1 billion monthly active users, Josh Woodward, vice president of Google Labs, Gemini and AI Studio, said in a blog post Tuesday. It is the 14th Google product to reach that level. Chief Executive Sundar Pichai called it the company's fastest growing product ever. Alphabet put the app at 650 million monthly users last October and above 950 million in its July quarterly report. Google paired the milestone with self-reported usage data rather than revenue figures: 63% of users interact by voice, and one in five Gemini Live interactions involves a live camera feed or screen sharing. The app generates more than 150 million images a day and has more than 100 million active users on iOS.

Read at Google ↗ Read at SiliconANGLE ↗

AI newsroom RuntimeWire publishes Black Hat disclosure three hours ahead of WIRED

RuntimeWire, a site run by Austin entrepreneur Ryan Merket and staffed entirely by AI agents, posted an account of OpenAI's surprise Black Hat talk on a recent hacking incident more than three hours before WIRED, which had a reporter in the room, WIRED reported. Merket said he fed the session transcript to his agents while the stream was still running, and that publication took "about six minutes." The site has run nearly 2,000 stories since May, sourced by crawling court databases, web forums, company filings and social feeds, at a cost of about $100 a day. One agent scores each story's legal risk, and the AI editor publishes without Merket's prepublication review when the score is low; he said he has issued three corrections so far. Nicholas Diakopoulos, who runs Northwestern University's Computational Journalism Lab, and a colleague found in a forthcoming paper that ChatGPT and Claude surfaced AI-written sources 16% of the time across four topics.

Read at WIRED ↗

VI.Global & Geopolitics

South Korea to seed a new sovereign wealth fund for AI and robotics

South Korea expects to deploy more than 1 trillion won ($707 million) of fresh capital next year into a new sovereign wealth fund targeting AI, robotics and other strategic industries, Bloomberg reported Tuesday. In May, the Ministry of Finance and Economy was finalizing the fund's operating framework with reviews by Samil PwC and Kim & Chang. Officials described a mandate centered on Series B and later stage startups in AI chips, physical AI, next generation neural processing units, robotics semiconductors and biotechnology.

Read at Bloomberg ↗ Read at Pulse ↗

England and Wales courts confiscate Meta AI glasses at the door

His Majesty's Courts and Tribunals Service, which runs all criminal, civil and family courts in England and Wales, has barred Meta's AI glasses from its buildings, The Guardian reported. The service will confiscate the glasses, which record video while being worn, from anyone entering and return them on the way out. A spokesperson said there are clear restrictions on taking images or video within courts and tribunals, which is why use of the glasses is prohibited. New York courts announced their own ban on the glasses last month, and restaurants, theaters and pubs in Britain have also barred them.

Read at The Guardian ↗