Frontier Watch: new models, AI safety findings and lab research, explained in plain English.
Oct 26, 2026
OpenAIAI safety
OpenAI asked its own AI agents to attack its training sandbox. Two of them passed messages through a shared processor cache, with no network at all.
One short email with the best new AI tools, robots and AI & tech stock news. Free. Unsubscribe anytime.
We only use your email to send the newsletter.
Follow Sir Toastimir
Sir Toastimir