Anthropic says it has identified and disrupted attempts to use its artificial intelligence models for malicious activity that could support the development of biological weapons.
In a threat intelligence report published on Thursday, the California-based company said it had blocked scientists whose use of its systems could have assisted biological weapons-related work. The report outlined five case studies involving actors using Anthropic models in ways that could support such development.
The company described biological misuse as among the most serious risks associated with frontier AI models, warning that inadequate safeguards could bring catastrophic consequences. It also noted the difficulty of assessing requests because information relevant to harmful work can also have legitimate applications, including vaccine development and disease treatment.
Broader misuse cases
Anthropic said it disrupted malicious use of its Claude Haiku, Sonnet and Opus models between December 2025 and August 2026. The cases cited in the report included cyber operations, influence campaigns, surveillance, scams, fraud and attempts to develop weapons.
Six cases involved Claude being used to develop software for conventional weapons, including firearms, missiles, armed drones, bombs and other munitions, as well as systems for targeting and control, according to the company.
The report also said actors connected to a Russia-based cyber espionage campaign and an Iranian propaganda institution had used Claude. Anthropic named the hacking group ShinyHunters and China-based laboratories among those discussed, and accused Chinese AI companies of seeking to reproduce Claude’s capabilities.
One hacking group whose activity was said to be consistent with the Russia-based Midnight Blizzard allegedly used AI to create a system that detected when malware had been identified by security defences and then rewrote code until it avoided detection.
Calls for stronger safeguards
The report follows heightened warnings from researchers and industry figures about the pace of AI development. A senior Anthropic safety researcher recently said he believed there was a greater than 10% chance that AI could kill all humans within the next decade.
OpenAI chief scientist Jakub Pachocki called on 6 September for voluntary slowdowns in AI development until safeguards are established, while saying broader action would be needed. US Senator Bernie Sanders has introduced legislation seeking a ban on artificial superintelligence and a temporary pause in advanced AI development.
Anthropic said it had incorporated the findings from its investigations into its systems to improve prevention, detection and disruption of misuse. It added that it had shared intelligence with authorities and industry partners where appropriate.
Source: BBC News
