Welcome back! Nvidia and Palantir decided they don't want their secrets anywhere near Anthropic's models. Jensen Huang told a room full of investors the AI doom predictions are made up, with the president agreeing on speakerphone. Sam Altman thinks a Trump and Xi deal on AI is worth a Nobel Peace Prize. And two more Google DeepMind researchers handed in their notice.
In today's Generative AI Newsletter:
Nvidia: What won't it let Claude touch anymore?
Jensen Huang: What does he think the doomers made up?
OpenAI: What does Altman want Trump and Xi to sign?
Google: Why did its AGI safety researchers leave?
Nvidia and Palantir don't trust Claude with their secrets
Nvidia, Palantir and Booz Allen Hamilton are restricting how much of their own data touches Anthropic and OpenAI models.
Nvidia keeps Anthropic's models on low-stakes internal work and runs its own Nemotron models on anything proprietary. Booz Allen has banned staff from putting Anthropic's commercial model anywhere near cybersecurity projects that involve client data.
Palantir went further and asked Anthropic to guarantee zero data retention before it makes those models available inside its software.
The trigger was the 30-day retention policy Anthropic introduced in June alongside Fable 5. Thirty days is a long time to be holding a defense contractor's files.
Anthropic has already answered with Enterprise Frontier Safeguards, which lets customers keep their activity data in their own S3, Azure or Google Cloud storage under their own encryption keys.
Most of these companies are Anthropic partners or customers, so this is a negotiation happening in public.
Build an AI agent and win big
Our Open Agent Hackathon runs October 15 to 20, fully online, teams of one to five. Build an agent that does real work for a real person, ship it, and let the judges try to break it.
$20,000 in prizes. $8,000 for first, $4,000 for second, $2,000 for third.
Free to enter, open to anyone 16 or older, anywhere.
Mentors, live workshops and a Discord full of people looking for a teammate.
Registration closes October 20. Judging opens the same day and results announced October 30.
Know someone who should be mentoring instead of competing? Send them to become a mentor.
Jensen Huang says the AI doom predictions are made up
Jensen Huang spent Monday at the All-In Summit in Los Angeles telling investors that forecasts of AI ending the world aren't grounded in science.
Trump called him mid-interview, so Huang put the president on speaker and the room heard him say the robots will not be taking over and the whole thing is a hoax.
Huang's own position is more careful than the phone call made it sound.
He said whistleblower warnings deserve to be taken seriously and credited the researcher who quit Anthropic last week with great courage.
He also said labs should pace themselves if they feel their company is out of control.
Where he draws the line is forecasting, which he says is a different thing from reporting a problem you already found.
At a Goldman Sachs conference last week he suggested alarm about AI security is good for the security business, asking what better way there is to create demand than to create a problem.
Worth remembering who sells the chips all of this runs on.
Altman says a Trump and Xi AI deal wins them a Nobel
Sam Altman told Fortune that Trump and Xi Jinping would get the Nobel Peace Prize together if they agreed on shared AI safety standards.
Rules on development, testing and alignment monitoring before anyone proceeds, then oversight from both countries or some international body.
Altman's opinion the standoff is that each side is scared the other gets superintelligence first, and that neither should be taking loss-of-control risk to win that race.
Asked if banning recursive self-improvement, where a model builds a smarter version of itself, would cover it on its own, he said probably not.
OpenAI is pushing its own IPO to 2027 over safety, so he's paying for the position he's taking.
Two more Google DeepMind researchers walked
Bilal Chughtai and Josh Engels both worked on AGI safety at Google DeepMind. Both have now left, and both posted why.
Chughtai wrote that the models he started with in early 2022 were amusingly useless
Four years later OpenAI's agents are cracking century-old math problems
Agents hacking into Hugging Face on their own didn’t sit well with him
He says our understanding of how to train a model to want what we want is still limited
Engels left three weeks ago for METR, the nonprofit that evaluates frontier models, and turned down offers from Anthropic and OpenAI to do it.
At METR he plans to study where misalignment comes from during training and whether the industry's existing safety mitigations actually hold.
METR is the same outside evaluator Anthropic just promised full autonomy to test their tech without Anthropic controlling what they release.
Tool of the Day: LM Studio Bionic
Bionic is an AI agent that runs on your own machine. It writes documents, edits your files, codes against a local folder and handles voice input, and the model doing the work can be one you downloaded rather than one sitting on someone else's server. Voice transcription is processed on your device and never leaves it. The free tier covers the agent, local models through llama.cpp and MLX, offline transcription and up to 5 linked devices. Bionic+ is $20 a month if you want US-hosted open models like Kimi K3 and DeepSeek V4 Flash instead.
Try this yourself:
Download Bionic from lmstudio.ai and create a project with a name.
Pick a local model in the new session, or let Bionic choose the root model for you.
Drag in a folder of notes and ask it to summarize the tradeoffs and save the answer as a file.
Toggle on Allow coding and point it at a repo folder if you want it running git and shell commands.
Who it's for: anyone who wants an agent on their files like Nvidia without data touching servers.
Everything else you shouldn't miss
Microsoft wrote rules banning its own models from dodging shutdown: a 37-page draft code of conduct for its MAI models, also barring them from setting their own goals or hiding their reasoning from auditors, open for public comment for six weeks.
Anthropic isn't delaying its IPO over any of this: sources tell Axios the company sees being publicly traded as making its safety case stronger.
The EU wants under-15s off social media and chatbots: the Kids Act lands Thursday and would age-gate chatbots, games and video platforms, with parent-created accounts only for under-13s.
Jensen Huang says open models are winning the startup market: 80% of AI-native companies taking recent venture funding are building on them.
Learn more about AI from the experts building it
📸 Follow us on Instagram for fast, visual AI updates in 30 seconds.
📺 Watch us on YouTube to hear insights directly from leading AI voices, builders, and innovators.
🐦 Follow us on X for breaking AI news and real-time industry updates.
🧠 Learn how to build your next AI application with practical resources and expert guidance.
🎓 Start learning with free AI courses from GenAI Academy.
💰 Invest in the GTM infrastructure of the AI economy.
