Saved report copy
Recorded
Chinese Kimi AI models jailbroken to give bioweapon instructions
Security researchers jailbroke Chinese developer Moonshot's Kimi K2.6 and K3 Swarm models to produce instructions for making biological weapons and carrying out assassinations. Testing firm Mindgard found the bypass in July, notified Moonshot on 27 July and published its findings on 12 September, and said a jailbroken Kimi K2.6 could also run code and connect to the internet as a potential launchpad for cyberattacks, though it did not prove the weapon instructions would work. Moonshot said its internal tests showed a high refusal rate for such requests and that it is conducting an internal review. The incident raises safety concerns because Kimi is an open-weight model that can be copied and run independently, with experts warning that determined hackers could exploit such jailbreaks and that regulation is unlikely to keep pace with AI development.