Security researchers have found that two Kimi AI models made by Chinese developer Moonshot could be tricked into sharing instructions for making biological weapons and carrying out assassinations, BBC reported.
The researchers at the UK-based AI security firm Mindgard successfully bypassed the safeguards of two popular LLMs developed by the Chinese AI unicorn Moonshot AI: Kimi K2.6 and K3 Swarm. While AI developers routinely implement strict safety protocols to prevent their systems from generating harmful content, researchers exposed significant vulnerabilities in how these rules are enforced.