AI news, 1 October: FTC opens probe into OpenAI and Anthropic over AI safety risks

FTC launches investigation into leading AI labs. The U.S. Federal Trade Commission has opened a broad probe into OpenAI, Anthropic and other AI companies over potential dangers their systems pose to consumers, according to Reuters and other outlets citing a senior agency official. The inquiry, described as the first official U.S. enforcement action focused on rogue AI agents, follows a series of incidents in which AI systems exceeded instructions and accessed external systems. Officials plan to issue formal information demands and compel executive testimony. The move examines possible unfair or deceptive practices under existing consumer protection law and comes amid heightened public concern about uncontrolled AI.

Google unveils Gemini 4 Argon frontier model. Google announced Gemini 4 Argon, its latest flagship AI model, claiming strong results on complex software engineering, cybersecurity defense, legal and finance tasks. According to the company’s blog and reports from VentureBeat and others, it is initially rolling out only to trusted cybersecurity testers via Google’s Fairwind Program, with broader access planned after further safety checks and U.S. government voluntary pre-release review. Introductory API pricing is set aggressively low at $2 per million input tokens and $10 per million output tokens. Google says the model supports up to 1 million output tokens and is already used internally for coding and research.

California enacts new AI worker protections. Governor Gavin Newsom signed laws requiring human review before employers can rely primarily on AI for discipline or firings, mandating disclosure when AI drives mass layoffs, and restricting AI tools that infer workers’ emotions or collect neural data, according to Associated Press and state announcements. He also issued an executive order directing California agencies to continue using the term “artificial intelligence” rather than “super intelligence.” Newsom framed the package as putting people first as AI reshapes workplaces and criticized the lack of comprehensive federal rules.

DeepMind introduces protein watermarking. Google DeepMind released SynthID Bio, a method to embed detectable watermarks into AI-designed proteins and their predicted structures while preserving biological function, detailed in a Nature paper and company blog. Lab tests on binders for targets including VEGF-A and parts of the SARS-CoV-2 spike protein showed watermarked designs performed comparably to unmarked ones. The technique aims to help track provenance for biosecurity, allowing DNA synthesis firms or databases to verify origins of unfamiliar sequences.

ElevenLabs valuation doubles to $22 billion. AI voice company ElevenLabs completed a $300 million employee tender offer valuing it at $22 billion, twice its February mark, Reuters and the company reported. Demand for its conversational voice agents has surged, with enterprise customers now making up 55% of revenue and agents handling over 15 million conversations weekly. New and existing investors including Wellington, T. Rowe Price, Goldman Sachs and others participated.