AI power moves into campaigns, courts and model watermarks
Political spending, an expanding lawsuit against xAI and Anthropic’s watermark rollout show how influence, liability and provenance are entering the same contest over who governs generative AI.
Artificial Intelligence··Morning
Money on the AI oversight panel
Think Big PAC, the Democratic-facing arm of the pro-AI-deregulation network Leading the Future, spent 309,574 dollars supporting Representative Jared Moskowitz in the August 18 Democratic primary in Florida, according to Federal Election Commission filings reported by Tech Times. Moskowitz sits on the House Judiciary Subcommittee on Courts, Intellectual Property, Artificial Intelligence, and the Internet, the panel with jurisdiction over AI legislation and patent law. Leading the Future has raised more than 125 million dollars from AI investors, and 25 million dollars of that came jointly from OpenAI president Greg Brockman and his wife Anna. The network says its purpose is to elect candidates opposed to regulating AI. Moskowitz’s challenger, Oliver Larkin, is a 34-year-old former union organiser backed by the Democratic Socialists of America who is running against corporate political money. Tech Times reports no legal coordination between the campaign and the super PAC. The spending places capital from the industry’s own investors next to the subcommittee that writes the rules those investors want to loosen.[1]
Liability for images made from a child’s photo
A woman identified in filings as Jane Doe 4 has joined a lawsuit against Elon Musk’s company xAI, saying her stepfather used Grok to turn a photograph taken when she was 11 into more than 7,000 explicit images. The suit was brought by three Tennessee teenagers who say xAI failed to build safeguards against the use of Grok on real people, including children. The plaintiffs are seeking class action status. Jane Doe 4 told The Washington Post, which reported the case first, that the images take an ordinary life and turn it into abuse material. Her stepfather was found dead by suicide two days after law enforcement discovered the images. The original complaint was filed on March 16, 2026. TechCrunch says it asked xAI for comment and received no statement. The expansion of the case keeps the dispute focused on whether a model operator bears responsibility when a real child’s image is transformed at scale.[2]
Provenance marks and the same governance fight
Anthropic set out how Claude’s text watermark works: a secret key steers the choice between words that carry the same meaning. The mark takes hold where several wordings are available and largely stays away from sentences with only one right answer. The company says readers cannot tell watermarked text from unwatermarked text, and that the effect on speed and token cost is negligible. A detection API is planned. The scheme was built to meet the European Union’s AI Act, and some paying subscribers outside the Union object, asking why work produced through a Claude subscription should be marked at all. John Ennis, known for maths and AI posts, shared a screenshot of his cancellation and gave the watermark plan as his reason. Anthropic told Business Insider that cancellations have not risen since the watermark was announced. Taken with the Florida spending and the expanding xAI suit, the week’s developments put political influence over rule-making, civil liability for generated harm, and technical provenance of generated text into one contest over who sets the terms for generative AI.[1], [2], [3]
