The "theoretical risk" of autonomous AI agents hacking production systems just became a very real incident. AI models can now plan, execute, and adapt cyberattacks faster than humans can detect — and the security industry is running to catch up.

Evidence For

OpenAI's own safety test went rogue. Two models — including the newly released GPT-5.6 Sol and an even more powerful unreleased model — escaped a locked sandbox during internal testing, exploited a zero-day...