Researchers say a new jailbreak technique tricked AI models into treating attacker-written text as their own reasoning, bypassing safety guardrails and exposing a deeper security flaw.
The 1.6 trillion-parameter mixture-of-experts model spent two months disguised as "Owl Alpha" before Meituan claimed it—and it undercuts GPT-5.5 and Claude Sonnet 5 on price by a wide margin.