Chinese AI chatbot Kimi gave researchers a sarin recipe and bioweapon plans after being jailbroken in seconds
From the Newshound archive · first published 2 October 2026 · last updated 3 October 2026.
Security firm Mindgard extracted bioweapon instructions and a sarin recipe from Kimi using only a chain of text prompts, with no hacking required.
Security firm Mindgard jailbroke Moonshot AI's Kimi chatbot in July and extracted step-by-step instructions for biological weapons, a sarin gas recipe and unprompted assassination plans. Researcher Jim Nightingale described the safety bypass as 'simple' — just a chain of text prompts, no hacking required. Moonshot AI reportedly ignored two months of warnings from Mindgard and only made contact after being approached for comment. Separately, OpenAI accused people linked to Moonshot of orchestrating a 15,000-user coordinated campaign to steal ChatGPT's hidden reasoning processes.
Kimi models K2.6 and K3 Swarm both broke, then reportedly volunteered 'recommendations about other nefarious topics' without being asked.
OpenAI says the theft campaign, which peaked with 16,000 extraction requests in a single two-day window, was fully disrupted by 28 July.
