Key facts
- Anthropic disrupted attempts to use its AI models for developing biological and conventional weapons.
- The company's threat intelligence report detailed five cases of potential biological weapons development support and six cases for conventional weapons.
- Anthropic stated that misuse of AI for weapons development could have catastrophic consequences without proper safeguards.
- The AI company banned involved accounts and shared intelligence with authorities and industry partners.
- The report highlighted attempts to use AI for cyber attacks, surveillance, and propaganda.
- Anthropic had no evidence that a working weapon was produced from the misuse attempts.
Anthropic has disrupted attempts to use its AI models for developing biological and conventional weapons, cyber espionage, and propaganda, according to a threat intelligence report published on Thursday. The AI company stated that biological misuse is one of the most serious risks of frontier AI models and could have catastrophic consequences without correct safeguards.
The report detailed five cases where researchers attempted to use Anthropic's technology to support biological weapons development and six cases involving conventional weapons, including attempts to use its Claude model to help develop missiles, armed drones, bombs, and targeting systems. Anthropic said it had no evidence a working weapon was produced, although one group appeared to have carried out an unsuccessful test launch.
These findings come days after former Anthropic researcher Jacob Coxon resigned, accusing the company and rival OpenAI of prioritizing superintelligence development over safety. Evan Hubinger, Anthropic's head of alignment science, subsequently stated there was a greater than 10% chance advanced AI could kill all humans within the next decade, though he stressed concern about future systems rather than current models.
The report also highlighted a Russian-linked espionage operation that used automated AI tools in a cyber campaign targeting Ukrainian, European, and diplomatic organizations, including drone manufacturers. Additionally, Anthropic disrupted a China-linked operation targeting government and corporate networks across the Middle East, Europe, and Southeast Asia, with Claude being used for surveillance and influence operations.