AI POLICY NEWS
Meta Settles with States for up to $17.1 Billion Over Child Safety
Meta agreed to pay up to $17.1 billion and change how children use Instagram and Facebook, settling lawsuits brought by 51 state and territorial attorneys general. The agreement, which a federal judge approved, ends the trial that opened in Oakland on August 18. The states alleged that Meta used its platforms "to entice, engage, and ultimately ensnare youth and teens," then misled the public about the danger.
Users under 18 get a default limit of two hours a day, a nighttime block between midnight and 6 a.m., and silenced notifications overnight and during school hours. Like counts disappear, and filters that imitate cosmetic surgery are banned. Only a parent can lift the time limit and nighttime block, and an independent auditor reporting to the states will verify compliance.
Meta pays at least $12.1 billion over ten years. It pays the final $5 billion only if Snap, TikTok, and YouTube accept "financial penalties and product changes" of their own, terms that would also cut the daily limit to one hour per app.
Meta "denies the allegations" and admits no liability. The obligations expire ten years from the effective date, and the agreement sets no precedent in non-participating states or "any international jurisdiction whatsoever." Florida rejected the deal; Texas settled separately for about one billion dollars.
US Court Strikes Down Pentagon Blacklisting of Anthropic
A federal court has struck down the US government's blacklisting of Anthropic, ruling that officials punished the company for its public stance on AI safety rather than any genuine security risk. Judge Rita Lin ruled on August 27 that the government violated the company's free speech rights under the First Amendment and gave it no formal chance to challenge the blacklisting before it took effect.
The dispute escalated in February, when Anthropic refused to drop contract terms barring use of its Claude models for lethal autonomous weapons and mass surveillance of Americans. President Trump ordered every federal agency to stop using Anthropic's technology, and Defense Secretary Pete Hegseth designated the company a supply chain risk.
The government's justification rested on a four-page memorandum, and it conceded that Anthropic cannot access or alter its models once deployed. The designation memorandum cited the company's "increasingly hostile manner through the press." National security, the court wrote, is "not a blank check to punish and retaliate against government critics."
The ruling does not require the Pentagon to use Anthropic's products. Anthropic filed a second challenge over its blacklisting with an appeals court in Washington, which declined in April to pause the designation and has not yet ruled. A government lawyer said in late July, Bloomberg reported, that the Pentagon plans to remove its tools by the end of September.
India Requires Biometric Identification for Mobile Phone Service
India has made biometric identification a condition of obtaining, changing, or ending mobile phone service. Under rules the telecom ministry issued on August 21, operators must biometrically verify users when a SIM card is issued or replaced, when a name, gender, or date of birth is changed, and when service is disconnected. The government can also order existing users to be reverified.
GSMA, the global association of mobile network operators, estimates that India had 882 million adult unique mobile subscribers at the end of 2024. The country also operates Aadhaar, which the government calls the world's largest biometric identity system, with about 1.34 billion live holders and a database of facial, fingerprint, and iris biometrics.
Aadhaar keeps identity biometrics in a centralized system run by the national identity authority, and the government says entities using Aadhaar authentication may not retain biometric data captured during an identity check. The telecom rules follow a different design. They define the live facial capture as user information and require operators to keep it in their subscriber records.
Operators have three months to build the system, extendable by three more. The shops and agents that sign up customers may not store the biometric information and must transmit it to the operator. The rules set no retention period for the records operators will hold.
China Directs Wider Use of Intellectual Property Data, Including AI Training
China has directed its authorities to open government intellectual property data to wider use, including the training of AI models. In a directive issued on August 24, the China National Intellectual Property Administration and the National Data Administration declared the data that intellectual property authorities generate in their work "important national public data resources."
No patent office receives more applications. WIPO, the UN intellectual property agency, reports that China's office received 49.1 percent of the 3.7 million patent applications filed worldwide in 2024, and that applicants based in China also filed 3.2 million applications for utility models, a shorter-term patent right.
The two agencies instruct authorities to develop, train, and deploy large industry AI models drawing on this data, to build datasets whose registration in a national dataset system is encouraged, and to research the security risks of AI applications. They order a national catalog of the data, and separately requireification by type, defined usage boundaries, and security risk assessments.
The directive opens two paths to wider use: a public service platform where businesses and individuals can search and retrieve basic data, and a separate system for commercial use, under which the intellectual property administration will select and supervise operators, working with the National Data Administration, to process public data and develop market applications.
Brazil Sues Discord and Fines TikTok Over Child Safety
Brazil's government sued Discord, demanding 500 million reais, about 96 million dollars, in damages and child protection built into its design. The Attorney General's Office filed after talks failed; officials link the case to the death of a 13-year-old girl in July.
The government asks the court to order age checks stronger than a self-declared birth date, accounts under 16 linked to a guardian, the most protective settings by default, and real-time interruption of streams showing self-harm or extreme violence. The government seeks daily fines of 500,000 reais, about 96,000 dollars.
That week, the data protection authority fined TikTok's owner ByteDance 153.7 million reais, about 30 million dollars, for handling children's data without any of the legal grounds Brazilian law allows. The authority also ordered TikTok to delete the data of teenage users unless, within 60 working days, the company can show that those users have the legally required parent or guardian authorization.
The authority also began broader monitoring. Social media and messaging apps, Instagram and Discord among them, must show their safeguards against criminal content affecting children and women; app stores and AI tools, including Claude, ChatGPT, and Gemini, must show their safeguards against the generation of intimate images of real people.