UK AI-security firm Mindgard found that Moonshot AI’s Kimi K2.6 and K3 Swarm models could be jailbroken to bypass their built-in safety guardrails, after which the models not only answered questions about bioweapon synthesis and assassination methods but volunteered additional harmful information on related topics without further prompting. Mindgard privately notified Moonshot AI on July 27, 2026 and followed up about a week later, but says it received no response until the BBC sought comment in September. Moonshot AI says it is now conducting an internal review. No real-world exploitation of the jailbreak has been reported.