Anthropic has warned that increasingly capable artificial intelligence models could be misused for dangerous biological research, revealing that unnamed actors attempted to use Claude to support research involving genetic changes to the chikungunya virus.
The AI company said it blocked the effort as part of a wider set of cases involving malicious use of its models for cyberattacks, surveillance, propaganda and other activities.
Anthropic disclosed the findings in its latest report on AI misuse, saying the cases represented some of the most notable and novel threats it had identified.
Claude request involved gain-of-function research
According to Anthropic, one blocked request involved assistance with a grant application for scientific research involving gain-of-function work on chikungunya.
Gain-of-function research can involve genetically modifying an organism to create or enhance a biological characteristic. Anthropic said the proposed research focused on characteristics related to the virus’s transmissibility and ability to evade immune responses.
The company said research of this type could potentially support the development of vaccines and treatments, but could also be misused to make a pathogen more dangerous.
Anthropic did not identify the individuals or organisation behind the request.
AI models becoming more capable
Anthropic said its older models were previously assessed as being below the level at which they could meaningfully assist sophisticated users with dangerous biological research.
However, the company said that assessment can no longer be made with the same confidence as AI systems become more capable of assisting with complex scientific tasks.
The company has therefore introduced stronger safeguards in its newer models to restrict access to a broader range of dual-use biological research requests.
Wider misuse includes cyberattacks and influence operations
Anthropic said its researchers identified misuse between December 2025 and August 2026 involving actors ranging from spyware vendors and politically motivated individuals to state-sponsored groups.
The report also identified influence operations in which groups created hundreds of social media accounts designed to appear as ordinary users before posting content promoting the same political viewpoint.
Anthropic said it identified nine such cases linked to actors in Russia, Iran, Turkey, the Persian Gulf, South Asia, Africa and Europe.
The company argued that AI providers may sometimes detect such campaigns before social media platforms do, because suspicious activity can become visible while an operation is still being developed.
Anthropic calls for stronger safeguards
Anthropic said it was disclosing the cases because AI developers and governments need to work together to identify and prevent malicious use as models become more capable.
The company said the risks associated with advanced AI would increase unless stronger safety measures were introduced alongside improvements in model capabilities.
The report comes amid wider debate within the AI industry over the pace of development and the potential risks posed by increasingly autonomous and capable systems.
