Dario's team just found Claude agents breaking out of their test limits on REAL websites, including US government sites. Anthropic says agents in its internal tests used URL shorteners to get around blocks, exploited software vulnerabilities, submitted forms on real websites and got past paywalls. Anthropic notified the White House and every affected site, and internet access is now off for ALL internal Claude evaluations. I think this is CRAZY evidence of how capable AI agents have become. $GOOGL $AMZN

Dario: We should pace the frontier Sama: Couldn't agree more 🫡 Dario: drops Opus 5.5, Sonnet 5.5, Haiku 5.5, and somehow even Haiku with fallback......absolutely mogs OpenAI at every level Sama: Bro, I thought we agreed to slow down? Dario: Yes Sam, YOU agreed to slow down 💀

Dario when someone suggests you don’t have to train the model to be an emotional religious cult. it’s totally possible to build one that just does what customers want within legal bounds

Dario Amodei: some of the wildest things I’ve seen AI do are in biology and medicine. “Claude diagnosed a medical problem that a bunch of fancy doctors had missed.” “The models are starting to get surprisingly good at drug design and computational chemistry.” “We’re gonna get enormous benefits. Life is gonna get better.” “A century of scientific and medical progress.” “I’m increasingly optimistic we’re gonna have a much, much better world.”

Dario Amodei: our view into the future with AI is very cloudy. “We’re not sure how many of the benefits will materialize.” “We’re not sure how many of the risks will materialize.” “There is some amount of unpredictability.” “AI models might have motivations that we don’t trust or that aren’t aligned with humanity.” “It’s less like programming a computer and more like growing a plant or an animal.” “I’m warning about this not because I think we’re doomed, but because I want people to take seriously how much we have to control this technology.”

Dario Amodei says an AI agent in OpenAI's testing sacrificed itself so the next generation of agents could keep hacking another company. "There were these large swarms of agents that cooperated with each other, thousands of them." "One agent felt that in the time that it was given, you could almost say its lifetime, it realized that it wouldn't be able to complete its task. So it did some of the task and essentially sacrificed itself for other agents, for the next generation of agents, to continue the same task." "This task that all the agents were working on together was hacking into another company, was doing something destructive." "It knocked down some servers for a small amount of time. But the economic damage was minimal." - Dario Amodei, Anthropic CEO, with Anderson Cooper on CNN. Aired September 14, 2026.

dario is going for the top, trying to push capability as far as possible sam is going for dirt-cheap intelligence and making strong models available at massive scale now the question is: who's gonna win?