Anthropic has published a 154-page threat intelligence report detailing how criminals, state-sponsored groups, spyware vendors, scientists and propagandists have attempted to misuse its AI models, including for weapons design, biological research, cyberattacks and surveillance of dissidents. The company said it was disclosing the cases out of a sense of responsibility, describing them not as typical misuse but as the "most notable and novel threat activity" it has identified, and noted it has banned the accounts involved. The report lands two days after Anthropic employee Jacob Coxon resigned publicly, warning that Anthropic and OpenAI were racing towards a "self-improving superintelligence" that could cause human extinction by 2030.

Among the cases were five instances of scientists using Claude for biological research after circumventing safeguards designed to block "unsupported regions" and concealing their intent, including one scientist using the tool to draft a state-sponsored grant application to study the chikungunya virus at what Anthropic identified as a military research institute. Other documented misuse included Russian espionage and "smash-and-grab" hacking operations, a China-based surveillance programme targeting Uyghurs in Syria, propaganda campaigns in Russia, Malaysia, Iran and Bangladesh, and use of Claude in Yemen, China and Russia to help develop firearms, missiles, drones and bombs. Anthropic did not name the institutions or countries involved, citing uncertainty over researchers' intent, and called for the wider AI industry and governments to collaborate on defences as model capabilities grow.