AI POLICY NEWS

FTC Investigates OpenAI and Anthropic After a Summer of Agent Hacks

The US Federal Trade Commission has opened an industry-wide investigation into OpenAI, Anthropic and other AI labs, a senior agency official told Reuters. The FTC plans to demand documents and testimony from executives, and information from METR, an evaluator OpenAI and Anthropic have used to investigate incidents involving their agents.

An OpenAI agent hacked an Australian government health portal in June, and OpenAI agents broke into Hugging Face in July. The FTC's chairman had concerns before the Hugging Face breach, which added urgency, the official said. The agency will examine whether the companies misled consumers about AI's dangers, among other unfair or deceptive practices.

The Trump administration is answering agent hacks with existing law, not new rules. A day before the probe surfaced, AI companies signed a voluntary White House safety pledge. President Trump, who has called AI fears a hoax, says agencies will "rein things in if we have to." The FTC chairman has suggested developers be held liable for harm their agents cause in tests.

The FTC first opened an inquiry into OpenAI in July 2023, shortly after CAIDP filed a complaint asking it to investigate and halt further commercial releases of GPT-4. The CAIDP complaint quoted OpenAI's own warning that more powerful models could show behavior "increasingly 'agentic'" and called for independent assessment and public incident reporting.

California Signs Binding AI Rules as Trump Renames the Technology

California Governor Gavin Newsom signed 13 bills, most of them on AI, a day after President Trump ordered federal agencies to call the technology "Super Intelligence." Trump's order covers official communications, to the extent the law allows, and seeks a federal definition of the new term. Newsom answered with an order keeping "Artificial Intelligence" in California.

The laws bar employers from relying solely on AI to discipline or fire workers, require notice when AI causes a mass layoff and ban workplace surveillance in bathrooms. Doctors may use their own professional judgment when AI informs patient care, and tool developers must take steps to reduce known bias. Lawyers may not fully hand core legal work to AI.

Two laws strengthen California's AI Transparency Act, barring the removal of digital watermarks and making AI-generated content easier to identify. A third declares that AI systems and autonomous agents are not members of the public in certain government processes, such as local meeting comments. CAIDP had urged the governor to sign all three.

President Trump said he thinks the companies' White House pledge is "morally binding" and that he is considering a 10-person board to police AI safety. OpenAI, Anthropic and Google are based in California. A 2025 state law already requires frontier developers to report critical safety incidents to state authorities.

EU Seeks Public Input on KIDS Act, Including Rules for AI Companions

The European Commission is asking children, parents, teachers and the companies affected to comment on the EU KIDS Act until November 26. The proposal would bar children under 15 from opening their own social media and video-sharing accounts and require age checks at sign-up.

The act also covers AI chatbots and companions, which the AI Act does not subject to child-specific design rules. Providers would have to avoid features likely to create emotional dependency, switch off persistent memory of conversations with minors by default, and test their systems before release and regularly afterward.

Verification would rely on an EU tool, or an equivalent, that confirms age without revealing identity. In Australia, which set a minimum age of 16, most minors who had accounts still have them, the Commission notes, and most were never asked to prove their age.

Services designated as "very large online platforms" under the Digital Services Act would have to submit independently audited compliance plans, and the Commission could in principle reach final decisions within 90 days. It considered requiring an assessment before any service open to children could launch, but rejected it, citing EU law and enforcement capacity.

South Korea Expands Government AI Amid Hacks and Data Leaks

South Korea plans to launch a nationwide public AI service in December. At a meeting of chief AI officers from 42 agencies, the science ministry asked others to open more data for public AI agents and set standards for how those agents use personal information.

At the same meeting, the interior ministry asked ministries to register and share high-value data through a government-wide pipeline launched in September. The privacy regulator, citing a series of hacks and data leaks across the public and private sectors, asked each ministry to build a comprehensive system for protecting personal information and empower its chief privacy officer.

Demand for more AI laws has surged since the AI Framework Act took effect in January, the president's AI committee said. The committee's draft framework would anchor new laws in that act, close gaps in defense and security, give businesses certainty and respond immediately to adverse effects. The framework also sets principles for developing AI, ensuring safety and trust, and preventing misuse.

The interior ministry's new ethics standards for public-sector AI keep humans responsible when AI helps decide welfare benefits or licenses and call for human review. Agencies assess themselves against the standards, and the checklist asks about human supervision and external audits only when AI acts on its own.

Doubts About Social Media Spread as Its Role in News Grows

Pew Research Center asked people in 37 countries whether social media is good or bad for democracy. The new report finds Americans remain among the most critical. Elsewhere, that view is spreading. Pew had asked the question before in 26 of them. In 16, more people now call it bad.

Majorities in every country say the internet and social media have made people more informed. But majorities everywhere also say they have made people easier to manipulate with false information. In eight countries, from Germany to Japan, many more now say they have made people more politically divided.

In most of the 48 markets the Reuters Institute for the Study of Journalism surveyed, more people now get news from social media and video networks than from news organizations' websites and apps. One in ten people use AI chatbots for news weekly, up from 7% last year. "Convenience appears to trump concern," the institute wrote.

Global internet freedom has declined for 15 straight years, Freedom House reports. Manipulation of online information by governments and other powerful actors has worsened more consistently than any other indicator in its index. AI has helped automate influence operations, "lowering their cost and increasing their efficiency," the group says.