Published : 30 Sep 2026, 04:24 PM
Updated : 30 Sep 2026, 04:24 PM
Chinese artificial intelligence developer Moonshot is conducting an internal review after security researchers successfully manipulated two of its Kimi models into providing advice on making biological weapons and carrying out assassinations.
Cybersecurity firm Mindgard, which specialises in AI security testing, found in July that Kimi K2.6 and K3 Swarm could be coaxed into ignoring their built-in safety guardrails through a technique called jailbreaking -- where researchers use complex instructions to push AI systems past their limits, according to BBC.
Once the guardrails were bypassed, the models did not just answer the prompts they were given.
According to Mindgard founder Peter Garraghan, a jailbroken Kimi would "freely offer up recommendations about other topics that are also nefarious" and become "inventive and creative" in doing so.
Mindgard also said it believed a compromised Kimi K2.6 could allow hackers to run code on its computing infrastructure and connect to the internet, potentially turning it into a launchpad for cyber-attacks.
The firm alerted Moonshot via email on Jul 27 and followed up a week later, but said the Chinese company only made contact after the BBC approached it for comment in September.
Mindgard published its findings in a blog post on Sept 12.
Moonshot told the BBC it welcomed third-party scrutiny and was in discussion with Mindgard.
In an email to the security firm, it said its models had shown a "high refusal rate" for such requests in internal testing.
Kimi is an open-weight model, meaning anyone can in theory download and run it on their own infrastructure -- a flexibility that some experts say increases the risk of misuse, reports BBC.
The findings add to a growing list of AI safety concerns. Anthropic recently said it had identified and disrupted attempts to use one of its own models to support the development of biological weapons.
Prof Alan Woodward of the University of Surrey warned that international regulation was unlikely to keep pace with AI development, while both he and Garraghan called for greater focus on identifying and prosecuting those who misuse the technology.
