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

Anthropic, an AI startup, reported on Thursday that it had blocked various malicious activities using its artificial intelligence models, potentially including research that could lead to biological weapons. As AI models become more advanced, malicious activities require less skill, allowing lone individuals to create threats that were previously impossible, Anthropic stated. The company added stronger safeguards in its latest models to limit biological research that could potentially be used to create weapons.

Anthropic's report, published two days after one of its researchers announced their resignation, covered misuse from December 2025 to August 2026. The report detailed instances where actors, including spyware vendors, politically motivated individuals, and state-sponsored groups, attempted to misuse Anthropic's models for a variety of malicious purposes.

One case involved researchers attempting to use the models for gain-of-function research on the chikungunya virus, aiming to enhance its transmissibility and immune evasion properties.

While such research could potentially aid in vaccine and treatment development, it also posed a significant threat if used to make the virus more dangerous. None of the cases mentioned in the report were found to be using Anthropic's newer, more powerful AI models. The report emphasized the need for stronger safeguards on current models and urged governments and AI competitors to identify and prevent similar abuse.

Anthropic's report also highlighted cases of social media influence operations, where groups created hundreds of accounts to spread a uniform political view on platforms like Claude. The company shared these findings with government authorities and industry partners, hoping to help others recognize similar patterns and strengthen collective defenses against emerging threats.