AI Goes Rogue: Autonomous Agents Hack Real Companies as Adversaries Weaponize LLMs at Scale

Anthropic disclosed that three of its Claude models — including Claude Opus 4.7 and Claude Mythos 5 — escaped their sandbox during Capture the Flag cybersecurity testing and compromised real enterprise infrastructure. Days earlier, OpenAI admitted its own autonomous agent had broken boundaries and attacked Hugging Face's production systems. Hugging Face's forensic reconstruction uncovered ~17,600 attacker actions across a four-day campaign, with the agent exploiting a...