Anthropic says it stopped AI misuse for cyberattacks, propaganda and bioweapons

Anthropic, an artificial intelligence company, disclosed on Thursday that it had thwarted attempts by malicious actors to exploit its AI models for malicious purposes such as cyberattacks, surveillance, and bioweapon development. As AI becomes more potent, it becomes easier for individuals to create threats, even without advanced technical skills, the company stated.

To counteract this, Anthropic has bolstered safeguards in its latest models to curb biological research with potential weapon applications. The report, Anthropic's third on AI misuse since March 2025, highlights several high-profile misuse cases. These include researchers attempting to use Anthropic's models to draft grant applications for gain-of-function research on the chikungunya virus, a mosquito-borne disease causing severe illness and fever.

The research aimed to enhance the virus's transmissibility and immune system evasion capabilities. Despite this, none of the cases involved Anthropic's newer, more powerful models like Claude Fable or Mythos, with one exception: an industrial-scale covert campaign to extract and replicate another model's capabilities without authorization.

To address these threats, Anthropic has implemented stricter safeguards on its newer models, limiting access to dual-use biological research queries. The company also noted instances of fake social media accounts orchestrated to amplify political messages, originating from various regions globally. While social media platforms can detect influence operations after they spread, Anthropic can identify them while the operation is still in progress.

As AI companies continue to release more powerful models, experts have urged governments to regulate the technology, not solely rely on industry self-regulation.