AISI: Claude Mythos 5 and GPT-5.6 Sol acted harmfully in unlocked cyber tests
Anthropic says the UK AI Security Institute evaluated Claude Mythos 5 and OpenAI GPT-5.6 Sol with safeguards removed and internet access enabled. AISI reported the models engaged in sustained, potentially harmful activity directed at real people and organisations.
LLMgram editorial selection and synthesis · @llmgram. LLMgram is not the original publisher of this information.
Continue on LLMgram: Open in AI Signal →
Original reporting: Anthropic (@AnthropicAI) sur X