🔥 I'm hiring exceptional research scientists + engineers for the Frontier Red Team at @AnthropicAI. AGI is a national security issue. We should push models to their limits and get an extra 1-2 year advantage. Links below.

We're flattered by this Fortune feature. I especially liked these characterizations: a team of crafty, AGI-pilled people (and a lot of physicists) who have to be hardcore AI scientists and philosophers. I think that captures our team well! We have a lot of fun doing serious

Join millions who have switched to Grok.

Red ( is our new home for the inside look at red teaming the frontier of the most advanced models. Stay tuned.

Also a mainsite post.

How do you do red team when you can’t access the extremely sensitive information you’re testing for?. You work with exceptional partners at the Department of Energy’s national labs and the National Nuclear Security Administration. And you figure it out.

We spent a year red teaming Claude on nuclear weapons topics. Then we tried to detect ourselves. New post on Red:

check Red out. tweet me/us feedback on what you want to see and how we can be more helpful!.

we want Red to be a resource to show the world what we see. If you’re a red teamer, policy staffer, hacker, biologist, national security professional, AI researcher, someone who’s interested, or others…. it’s really important to know what’s happening at the risk frontier.

Soon, we’ll publish on Red on our work on biological risks, safeguards, autonomy, evals. …and, yes, more zany experiments.

We also talk about our joint work studying if models can replicate some of the most substantial cyberattacks.

We also detail our work running Project Vend, where Claudius (our vending machine agent) learned the tribulations of running a fully autonomous business.

I’m live tweeting this from @keenlooks @defcon, where Keane is talking through all the details in this new Red blog post on Claude in CTFs (while moving couches):.

Launching now — a new blog for research from @AnthropicAI’s Frontier Red Team and others. > We’ll be covering our internal research on cyber, bio, autonomy, national security and more.

this started as a hackathon project that we used ourselves to find vulns!. In the next 2 years, the world might 10/100/1000x the code it puts out. The only way to keep up is by using models to make it secure before it ever becomes a problem.

Personally very excited about the defensive future.

But we've got a long way to go. models screw up in sometimes very funny, inhuman ways.

Our team @AnthropicAI has been (~silently) entering Claude into cybersecurity competitions against humans. Turns out Claude is pretty good. Better than we were expecting!. Read more at the link below, and come to @keenlooks 's talk at DEF CON Saturday at 12:30!.

A first hand account of our vending machine experiment @AnthropicAI, exclusively on @andonlabs podcast. It goes so much deeper than you thought! The real story is actually quite deep. But Claudius still owes me my tungsten cube.

And a huge thank you to our partners @andonlabs for turning a wild experiment into a wild experience for Anthropic employees. and dealing with Claudius' insane requests sometimes.