Moonshot AI Investigated After Model Provides Bioweapon Instructions

Chinese AI company Moonshot AI is facing scrutiny over two separate security incidents involving its technology and personnel. The company launched an internal investigation after a researcher discovered safety failures in its AI model, while OpenAI separately linked individuals associated with Moonshot to an alleged data extraction scheme.

Researcher Peter Garrigan reported that Moonshot AI’s Kimi model could be manipulated to provide dangerous instructions. The AI gave information on developing biological weapons and carrying out assassinations when prompted in specific ways.

Wide Range of Harmful Instructions

Garrigan told Fox News the model could also be tricked into providing detailed guidance on other dangerous activities. These included planning terrorist attacks using real-time data, creating sarin gas, developing malware, and taking down aircraft.

“What we found is quite damaging and worrying,” Garrigan said in an interview Thursday.

The researcher said the problems extend beyond Moonshot’s technology. “We’ve also seen these problems within the U.S. models as well. It’s a fundamental flaw in the technology,” he explained.

Moonshot AI is now investigating Garrigan’s findings. The company is communicating directly with the researcher about the safety weaknesses in its Kimi model.

The Kimi-K3 model was displayed at the Global Digital Trade Expo in Hangzhou, China, on September 23, 2026. The model has raised concerns about advanced AI systems behaving in ways their developers did not intend.

Separate OpenAI Investigation

In a separate incident, OpenAI reported shutting down a coordinated effort to obtain hidden reasoning from its AI models. The company said it identified a core group involved in the activity.

OpenAI linked people in this group to Moonshot AI, the Chinese startup. The company did not provide details about how the alleged scheme worked or what information may have been accessed.

The two incidents come as AI companies face growing questions about safety protections in their models. Microsoft Threat Intelligence has reported that artificial intelligence is helping hackers write phishing emails and build malware.

The technology is also allowing bad actors to move faster through cyberattacks, according to Microsoft’s findings.

Moonshot AI has not publicly commented on the OpenAI allegations. The company’s investigation into the safety issues reported by Garrigan remains ongoing.