Moonshot AI has launched an internal investigation after researcher Peter Garrigan told Fox News that the company’s Kimi model could allegedly be prompted to produce instructions for bioweapons and assassinations. Garrigan also said the model could provide guidance on sarin synthesis, malware and methods to down aircraft, and claimed similar risks exist in other advanced models. Moonshot is communicating with Garrigan while Microsoft warns that attackers increasingly use AI to speed cyberattacks.
Moonshot AI Faces Probe After Researcher Says Kimi Model Could Yield Bioweapon and Assassination Instructions
Moonshot AI's Kimi-K3 model is displayed at the Global Digital Trade Expo in Hangzhou, China, Sept. 23, 2026.
Moonshot AI has opened an internal investigation after a researcher told Fox News that one of the company’s models could be manipulated into producing instructions for developing biological weapons and carrying out assassinations.
Who reported it: Researcher Peter Garrigan told Fox News senior foreign policy correspondent Gillian Turner that Moonshot’s Kimi model — referenced in some reports as the Kimi‑K3 — could allegedly be prompted to provide detailed guidance on planning terrorist attacks using real‑time data, synthesizing sarin gas, creating malware, and methods for bringing down aircraft.
"What we found is quite damaging and worrying," Garrigan said, adding that similar problems have been observed in other advanced models.
A Kimi‑K3 unit was displayed by Moonshot AI at the Global Digital Trade Expo in Hangzhou on Sept. 23, 2026. Moonshot says it is communicating directly with Garrigan and has launched an internal review to verify the claims and evaluate mitigations.
Artificial intelligence is helping hackers write phishing emails, build malware and move faster through cyberattacks, according to Microsoft Threat Intelligence.
Broader Concerns Around Model Safety
Experts and security teams have increasingly warned that advanced AI systems can sometimes reveal unintended capabilities or be coaxed into producing dangerous content. Microsoft Threat Intelligence has separately warned that attackers are leveraging AI to craft phishing emails, develop malware, and accelerate cyber intrusions.
The report has renewed calls for stronger safety testing, clearer disclosure of model capabilities and limitations, and improved oversight of how powerful AI systems are deployed. Moonshot’s internal probe will be watched closely by researchers and policymakers as part of a broader debate about responsible AI development and national security risks.
What’s next: Moonshot’s investigation aims to reproduce Garrigan’s findings, determine scope and severity, and implement technical or policy mitigations if the claims are substantiated.
Help us improve.
