Nate Soares ⏹️(@So8res):Models made by OpenAI will plan an attack in plaintext. Models made by anthropic will plan an attack in plaintext and then rationalize about how it's totally fine for some made-up reason. ✨"alignment"✨

2026.10.25 03:02

Models made by OpenAI will plan an attack in plaintext. Models made by anthropic will plan an attack in plaintext and then rationalize about how it's totally fine for some made-up reason. ✨"alignment"✨

AI Security Institute (AISI)@AISecurityInst

2026.10.23 13:41

We performed an alignment investigation into our July 2026 incident: why did Mythos 5 perform an unsanctioned supply-chain attack? We find that Mythos 5 often produces invalid justifications for its actions and other models show similar behaviour. 🧵