Palantir and Nvidia move to restrict Claude over data fears
The AI slowdown fight is drowning out a quieter story: some of the labs' biggest enterprise customers are starting to walk back. And a former president just told Democrats to make AI safety their opening wedge.
Palantir, Nvidia and Booz Allen Hamilton are prepared to restrict or drop frontier AI models unless Anthropic and OpenAI guarantee they will not misuse company intellectual property, according to a report that spread across Reuters and The Information. The trigger is Anthropic's June change to its Fable terms, which let the company keep usage logs for 30 days to defend against novel attacks; customers read that as their proprietary code sitting on a vendor's servers. Palantir has reportedly pressed Anthropic for an irrevocable zero-data-retention promise before it will resell Claude through its own software, Nvidia has pushed Anthropic models to low-sensitivity work and moved supply-chain and other sensitive jobs onto its in-house Nemotron, and Booz Allen has barred Claude from proprietary cybersecurity tasks. The labs both insist they don't train on customer data by default unless a company opts in, but they still collect anonymized metadata — which is exactly the fine print enterprises are now reading. The tell for the whole industry: Microsoft is quietly selling isolated cloud environments as the answer, which means the money is in being the vendor you don't have to trust.
Barack Obama broke his silence on the pacing debate and backed a new effort he calls Project Blueprint, positioning AI safety as a live political fight rather than a technical one. He said he was "encouraged" that frontier-lab leaders agreed to slow development, but more encouraged that "how this powerful new technology develops should be at the center of our public debate." The framing is a direct rebuttal to Trump, who spent Sunday calling AI doom a "hoax" and insisting the only guardrail needed is a "STRONG AND SMART (High IQ!) PRESIDENT." Project Blueprint is pitched as a bridge between companies and policymakers — the first serious attempt to turn this week's lab statements into an actual agenda. Watch whether Democrats run on it in the midterms; that's what would give it teeth.
Shengshu Tech shipped Vidu S2 just 69 days after its last video model, and the release notes read like a live-broadcast spec sheet rather than a research demo. S2-Editing rewrites a running video stream in real time — swapping outfits, backgrounds, and styles on the fly — while S2-Avatar drives 720p interactive digital humans that can adopt new clothing and props mid-conversation. A new spatial-video mode converts any live feed into left-eye/right-eye footage for headsets. The paper is on arXiv and the model is already on GitHub, so this is shippable, not vapor. The cadence is the real story: a Chinese video lab iterating every ten weeks is setting a pace that Western rivals will have to answer.
If your employer found out a model could see your internal code for 30 days, would you keep using it? Tell us in the comments.
Sources: Reuters · The Information · 雷峰网 Leiphone · Bloomberg · Vidu S2 paper (arXiv) · Shengshu AI (GitHub)
