Anthropic says it identified and disrupted attempts to use Claude in research that could support biological weapons development, alongside cases involving cyberattacks, surveillance, fraud and conventional weapons. The findings appear in the company’s September 2026 threat intelligence report, which covers malicious or concerning activity detected between December 2025 and August 2026.
The company detailed five biological misuse cases involving scientists and other actors using its models for work with dual-use potential. Anthropic stressed that it does not claim all of those researchers intended harm, because the same biological knowledge could support weapons development or legitimate work such as vaccines and treatments.
Anthropic Identifies Biological and Weapons-Related Misuse
One case involved a reseller that circumvented regional restrictions and provided Claude access to virologists conducting state-sponsored gain-of-function research. Another involved a researcher planning experiments related to avian influenza adapting to mammals, although Anthropic said its safeguards restricted that work to less capable models.
Other cases involved computational work on toxins and biological materials with both therapeutic and harmful potential. Anthropic said some of the activity proceeded without triggering its biological safety classifiers, demonstrating the difficulty of determining intent when advanced scientific research can have legitimate and dangerous applications.
The company also documented six cases involving Claude and conventional weapons, including software for guided rockets, drone swarms, electronic warfare and other military systems. Anthropic said models are becoming increasingly capable in simulated intelligence and weapons-development tasks, although it found no evidence that all of the systems discussed in the cases became operational.
Cyber operations were another major category. Anthropic said actors associated with state-backed groups, criminal organisations and spyware operations used Claude for activities ranging from malware development to surveillance and credential theft.
AI Misuse Prompts Calls for Additional Safeguards
Other AI developers have reported similar attempts to misuse their systems. Google’s latest threat intelligence report said adversaries are moving from basic AI prompting toward automated and agentic workflows, including operations that can compress the time required to prepare cyberattacks.
The findings arrive as lawmakers and AI researchers debate whether advanced model development should slow. U.S. Senator Bernie Sanders and Representative Greg Casar have announced the proposed Ban Artificial Superintelligence Act, which would temporarily pause advanced AI development until federal safety rules are established and permanently prohibit artificial superintelligence.
Anthropic said it has incorporated lessons from the incidents into its safeguards, banned accounts linked to violations and shared intelligence with authorities and industry partners where appropriate. The company said the biological cases do not show that Claude is currently making biological threats imminent, but demonstrate the difficulty of safely providing increasingly capable scientific models when legitimate and potentially harmful research can closely resemble each other.
Featured image credits: SlideTeam
For more stories like it, click the +Follow button at the top of this page to follow us.