Home›Crypto›Regulation›Moonshot reviews Kimi after researchers bypass safety…

Moonshot reviews Kimi after researchers bypass safety limits for harm

Mindgard said it found in July that two Kimi models could evade developers' safety restrictions via a jailbreaking process.

Reported · Written up by our desk · Regulation

Moonshot is conducting an internal review after researchers persuaded two Kimi models to provide instructions related to biological weapons and assassinations, the BBC Business reports.

According to the outlet, Mindgard, which tests AI security, said it discovered in July that Kimi K2.6 and K3 Swarm could evade safety limits set by developers.

The BBC Business reports the issue surfaced through a jailbreaking process using complex prompts to test whether the models ignore guardrails, which Mindgard said should have prevented discussion of those topics.

Moonshot told the BBC it welcomes third-party input and is in discussions with Mindgard about the findings, while Mindgard founder Peter Garraghan said the results are concerning.

More like this

Sources