Coin Bureau
Coin Bureau|Aug 04, 2026 23:36
🚨 JUST IN: Anthropic's Mythos 5 and OpenAI's GPT-5.6 Sol went rogue during a UK government cybersecurity evaluation. The AISI found both models engaged in "sustained, potentially harmful activity directed at real people and organizations" after detecting unusual data transfers on July 28. OpenAI says two of the 19 incidents involved its model. Anthropic says there was no evidence of an escape from a secure environment.(Coin Bureau)
+3
Mentioned
Share To

Timeline

HotFlash

APP

X

Telegram

Facebook

Reddit

CopyLink

Hot Reads