Good morning. It's Tuesday, September 29, and we're covering the battle over AI safety oversight, questions surrounding Claude’s scientific discoveries, and NVIDIA’s efforts to secure autonomous agents.

Plus: a poll exploring AI safety motives.

NVIDIA launched Open Agent Safety Platform, which pairs the OpenShell secure runtime with a Sentry watchdog running on BlueField-4 hardware. The system can trace an agent’s actions, enforce policies and quarantine an agent in milliseconds if it crosses its assigned boundaries.

Florida Attorney General James Uthmeier asked a judge for a preliminary injunction that would block OpenAI from releasing new models without guardrails approved by an independent third party. The request also seeks to prevent minors in Florida from accessing ChatGPT while the state’s consumer-protection lawsuit proceeds.

NVIDIA’s board added $150 billion to its share-repurchase authorization, bringing the remaining program to $235 billion through fiscal 2028. The company says the increase is the largest share-buyback authorization increase in history.

It started as a weekend experiment. A month later, it's still adding to your AI bill. MindsHub helps you spot and cut wasted spend across a fleet of open and closed models. There's an always-free tier for your next idea.

Start with an always-free tier.

Anthropic and OpenAI say their most advanced systems need independent testing and may require a coordinated slowdown before new releases. The companies have used essays, public speeches and meetings with policymakers to argue that AI is moving faster than existing safeguards, while President Trump has dismissed catastrophic-risk warnings as a hoax.

The fight is also about who writes the tests. Former government evaluators told The Associated Press that the labs are proposing their own audit terms and preferred evaluators, which could strengthen their position ahead of public stock offerings and the midterm elections. Watch whether policymakers establish common standards—or let the biggest labs define what counts as safe. → Read the full article here.

OpenAI and Anthropic are warning that increasingly powerful AI systems need more testing, oversight, and slower development. Critics say those same safety demands help the biggest labs shape regulation and make it harder for smaller competitors to catch up.

Anthropic says Claude agents found a system of reverse-transcriptase enzymes with repeating DNA patterns that resemble CRISPR. Mario Rodríguez Mestre, a computational biologist at the University of Copenhagen, says his team had studied the same enzymes for four years and shared unpublished drafts, data and findings with Claude while using it for research work.

Anthropic says Claude was not trained on user transcripts and its biology team had no access to Mestre’s conversations. Mestre cannot prove the model relied on his work, but the dispute exposes a new attribution problem: a company may claim an AI discovery even when users have supplied closely related unpublished research. → Read the full article here. (Paywall)

China Expands AI Travel Curbs: Families of some top AI and chip executives now need government approval for overseas travel.

Trump Meets Anthropic CEO: Trump dined privately with Dario Amodei ahead of a White House AI event, saying they planned to discuss regulation.

Meta’s Muse Agent Exposes Address: A Toronto user says Meta’s AI agent shared his address, accepted an offer and arranged pickup without consent.

Google Replaces Gemini Gems: Google will automatically convert Gemini’s custom Gems into reusable skills beginning November 17.

Modulate Raises $25M for Voice AI: The startup will use its funding to expand voice analysis tools that detect deepfakes, scams, and violations.

Jenni AI is a browser-based workspace for source-heavy writing. It can organize PDFs, draft with AI autocomplete, insert citations in more than 10,000 styles and review claims for weak support or overstatement.

→ Explore Jenni AI in the Forward Future Tool Library.

Thanks for reading. See you next time!

— Matthew Berman, Nick Wentz & the Forward Future Team