AI company Anthropic disclosed this week that it disrupted five separate attempts by scientists to use its Claude models for research that could support biological weapons development, including a gain-of-function study intended for a military research institute, as part of a broader threat intelligence report covering misuse of its technology.

The report, titled "Detecting and Countering Misuse of AI: September 2026," catalogs activity the company identified and shut down between December 2025 and August 2026 across seven categories of harm, according to Tech Times' coverage of the findings.

Blocked, But Not Everywhere

Anthropic said its biological safety systems successfully intercepted direct requests for bioweapons construction guidance. But the company also acknowledged a gap: researchers operating in regions without full access to its tools routed dual-use scientific queries — covering pathogens, immune evasion and other high-risk biological topics — through other, less-guarded systems instead.

A GAIN-OF-FUNCTION STUDY INTENDED FOR A MILITARY RESEARCH INSTITUTE

On the cyber side, the report documents a Russian state-linked espionage campaign against Ukraine that ran for 130 days, according to the Washington Times. The operation targeted 24 of 27 Ukrainian institutions it went after, including government ministries, defense bodies and manufacturers in the drone supply chain.

A Longer List

Beyond weapons research and state-backed hacking, Anthropic's report also flags an Iranian propaganda operation traced to a state-linked institution, a network of fake dating apps built to defraud users, hotel Wi-Fi scam infrastructure, and attempted hacking by Chinese competitors aimed at extracting the capabilities of Anthropic's own models. The company said it is sharing relevant findings with government and industry partners to help harden defenses elsewhere.

Why It Matters

The disclosures land amid a broader industry reckoning over how quickly increasingly capable AI models can be weaponized by bad actors, and how much responsibility falls on the companies building them to police misuse after the fact rather than prevent it at the model level. Rival labs including OpenAI and Google have published similar threat reports in recent months, part of an emerging norm of public disclosure that AI companies have adopted partly in response to regulatory pressure.

What's Next

Anthropic said it continues to refine detection systems to catch dual-use research being laundered through less-monitored platforms, and that it expects future reports to track whether the proxy-routing workaround identified in this cycle becomes a more widespread pattern across the industry.