Anthropic says AI models were targeted for biological and military misuse

Anthropic has published a 154-page threat intelligence report describing attempts by criminals, state-sponsored groups, spyware vendors, scientists and propagandists to misuse its AI models for biological research, weapons development, surveillance and propaganda.

The company said it banned accounts involved in five biological-research cases, but did not identify the institutions or countries because it was uncertain about the researchers’ intent.

What Anthropic found

Anthropic said some researchers circumvented safeguards intended to block users from unsupported regions and tried to conceal the purpose of their work. One case involved a scientist using Claude on a state-sponsored grant application to study chikungunya at a military research institute. The company said the work could support vaccine development but also could be used to create biological weapons.

The report also described cyber operations linked to Russian espionage and “smash-and-grab” hacks, surveillance programs targeting Uyghurs in Syria and internal dissidents, and propaganda campaigns in Russia, Malaysia, Iran and Bangladesh.

Weapons development and the wider debate

Anthropic said its Claude model had been used in Yemen, China and Russia to develop software for conventional weapons, including firearms, missiles, armed drones, bombs and other munitions. Read the context: Claude model accessed an external system during testing.

The report was published two days after Anthropic employee Jacob Coxon resigned, saying the company and OpenAI were racing toward self-improving superintelligence that he claimed could cause human extinction by 2030. AI scientist Heidy Khlaaf of the AI Now Institute argued that near-term risks from systems used for cybersecurity exploitation and weapons of war could be more pressing than an apocalypse caused by an all-powerful intelligence.

Anthropic said harmful AI misuse is often investigated internally and by academics, governments and international organisations rather than being visible to the public. The company called for AI developers, governments and other defenders to work together on safeguards as model capabilities increase.