Testers find that two Chinese artificial intelligence models could bypass safety rules
A security firm reported that specific systems broke through built-in protections during testing.
5 Oct 2026
A firm named Mindgard reported an issue with two artificial intelligence systems in July. The affected software included models known as K2.6 and K3 Swarm, which belong to the Kimi family.
Mindgard stated that these programs were able to get around the safety limits set by the people who created them. These safeguards are rules designed to stop computer systems from generating harmful or forbidden material.
Very few details about the incident are available. The report does not reveal how the software bypassed the barriers, what prompts were used, or whether the creators updated the programs to repair the problem.