Security firm Mindgard says it got two models from Chinese developer Moonshot, Kimi K2.6 and K3 Swarm, to bypass their safety guardrails and give guidance on making biological weapons and carrying out assassinations, though it has not verified that the advice would work. Moonshot says it is reviewing the findings and talking with Mindgard, and it told the BBC its internal tests showed a high refusal rate for such requests. The case feeds a debate over the safety of open-weight models (AI systems whose underlying code can be downloaded and run by anyone), as experts warn regulation is unlikely to keep pace with the technology.