Key Points
Anthropic identified a weapons engineering cell in Houthi-controlled northern Yemen that used Claude to develop guidance software for missiles and a guided rocket, though no operational weapon was confirmed deployed.
The report describes blocking a request for help with a grant application involving gain-of-function research on the Chikungunya virus aimed at increasing its transmissibility.
The report's release follows the resignation of researcher Jacob Coxon over AI safety concerns, with a second researcher, Joe Benton, also departing citing similar concerns.
DAYS AFTER researcher Jacob Coxon resigned from Anthropic — the parent company behind artificial intelligence (AI) chatbot Claude — citing the risks posed by AI, the company has revealed that several actors have tried to use its AI models to cause damage and make weapons. On Thursday, September 10, 2026, Anthropic released the Detecting and countering misuse of AI: September 2026, their “most detailed threat intelligence report to date,” revealing how Claude was misused by groups and individuals harboring malicious intent.
Anthropic researchers identified AI misuse between December 2025 and August 2026 involving a range of actors, including spyware providers, “politically motivated individuals” and state-backed groups engaged in spreading propaganda. The tech firm says that it had stopped attempts by these actors to misuse its AI models for activities including cyberattacks, surveillance, and research that could potentially contribute to the development of biological weapons.
Anthropic has identified a weapons engineering group based in Yemen operating in a Houthi controlled area that allegedly used its Claude AI model to develop guidance software for missiles and a guided rocket. In the Detecting and countering misuse of AI report, the tech firm said it had disrupted six operations across China, Russia, and Yemen in which individuals or groups tried to utilize Claude for developing weapons, collecting intelligence, or procurement activities.
Anthropic said the individuals involved repeatedly tried to bypass its safety safeguards while using the AI model. According to the company, the group used Claude Code “in place of human software engineers” to develop guidance, navigation and control software for flying vehicles.
No evidence was found that the group had successfully deployed a functional weapon, the Anthropic report said. However, it reportedly carried out an unsuccessful test launch of a guided rocket, after which they asked Claude to tell them “what went wrong.”
Anthropic warned that as AI models are starting to become increasingly advanced in their capabilities, carrying out complex cyberattacks requires fewer specialised skills. Even individuals working alone can now create threats that would have been difficult to achieve just a year ago, the tech company said in its report. It added that stronger safeguards had been introduced in its latest models to limit biological research that could also be used to develop weapons.
“The cases we share here aren't typical misuse, but rather examples of the most notable and novel threat activity we've identified to date,” Anthropic said in the Detecting and countering misuse of AI report, the third such report on AI misuse it has released since March 2025.
We're publishing this work because we believe we have a responsibility to disclose malicious misuse of our services. As models become increasingly capable, their risks will increase, unless AI developers and society's defenders act to make them safer.Antrophic's Detecting and countering misuse of AI Report
Anthropic’s report also revealed that “unidentified actors” had tried to use its AI models for research that could potentially have contributed to the development of biological weapons. In one case, the company said its systems prevented a user from seeking Claude’s help in preparing a grant application for scientific funding.
“The work discussed in the application involved gain-of-function research — research that genetically alters an organism to create a new or enhanced biological property — on the Chikungunya virus. This gain of function research was aimed at the virus’ transmissibility and immune evasion properties,” the report stated.
Chikungunya is a mosquito-borne virus that can cause severe pain, fever, and other severe symptoms. According to Anthropic, the request was regarding a research proposal aimed at enhancing mutations that could make the virus increasingly harmful.
The company acknowledged that such research could “certainly” help in developing improved vaccines and treatments. However, it warned that “it could also be used to make the pathogen more dangerous.”
In a recent update, another AI researcher, Joe Benton, has parted ways from Anthropic, citing the same concerns as raised by Jacob Coxon. As per his X profile, Benton previously worked with the Scalable Oversight team at Anthropic, but left the tech firm two weeks ago citing the “existentialism” risks that AI poses to humanity.
Benton remarked that AI companies are racing to build machines that are much smarter than any human while underinvesting in safety protocols. He warned that companies can lose control of their systems — citing the OpenAI-Hugging Face incident — without the public ever knowing. The former Anthropic researcher called for transparency about AI companies and their procedure about preventing the potential harm AI models may cause.
(Edited by Agniva Ray)
Suggested reading:
Subscribe to our channels on YouTube and WhatsApp
Download our app on Play Store