Anthropic says it blocked misuse of its AI that could have supported biological weapons

Anthropic, an artificial intelligence startup, has blocked attempts by malicious actors to utilize its AI models for harmful purposes, including cyberattacks, surveillance, and research potentially leading to biological weapons. As AI models become increasingly powerful, they can now create threats that were previously impossible, according to the company.

To address this, Anthropic has implemented stronger safeguards within its latest models to restrict research that could be used for biological weapons. The company published a report detailing instances of misuse, including a request to use its AI to support the research of a virus to enhance its transmissibility and immune evasion properties.

This gain-of-function research involved a chikungunya virus, which causes severe symptoms and disabling effects. While such research could lead to beneficial advancements, Anthropic warned it could also be used to create a more dangerous pathogen. The report reveals how malicious actors, ranging from spyware vendors to state-sponsored groups, abused Anthropic's AI models.

One case involved an industrial-scale campaign to extract and replicate the model's capabilities without authorization. Anthropic has applied stronger safeguards to its more recent models, such as Claude Fable 5, to restrict access to a wide range of dual-use biological research queries.