AI company Anthropic said it has disrupted several attempts to use its Claude models for research that could potentially support the development of biological weapons, amid rising concerns over the risks posed by increasingly capable artificial intelligence.In its latest threat intelligence report released Thursday, the US-based company said it identified five cases involving users who were using its AI models in ways that could support biological weapons development.Anthropic said it could not determine with certainty whether the research was intended for malicious purposes, but chose to intervene because the same scientific knowledge can have both medical and military applications.
AI used to explore dangerous biological research
In one case, reported in May, a scientist sought help with a grant application to engineer mutations in chikungunya, a mosquito-borne virus, with the aim of making it more harmful in live animals, according to Anthropic.The company said the case raised additional concerns because it determined that the research was intended for a military research institute. Anthropic subsequently banned the accounts involved.“What we don’t know is if the research was meant to be weaponised,” Jacob Klein, Anthropic’s head of threat intelligence, told The New York Times.Anthropic said the difficulty lies in distinguishing legitimate scientific research from work that could eventually be used for harmful purposes.Research into pathogens and other biological systems can potentially contribute to vaccines and treatments while also providing knowledge that could be misused to develop dangerous biological agents.The company described biological misuse as one of the most serious risks associated with frontier AI models and said it had shared relevant intelligence with authorities and industry partners.
Claude used in cyber, weapons and espionage operations
The report also detailed other attempts to misuse Anthropic’s AI models, ranging from cyberattacks and scams to surveillance and the development of conventional weapons.Anthropic said Claude had been used in efforts linked to missile projects, armed drones, bombs and other weapons, as well as systems used for targeting and control.The report also identified suspected state-backed hackers and criminal groups that used Claude in cyber operations. In one case, a group allegedly linked to Russia used AI to develop malware that could detect when it was being flagged by security systems and modify its code to evade detection.Anthropic said its models had also been used in espionage and influence operations, including activity linked to Russia and Iran.The company said the cases detected between December 2025 and August 2026 involved its Claude Haiku, Sonnet and Opus models. It said it had disrupted the activity and incorporated its findings into systems designed to better detect and prevent similar misuse.
Rising concern over AI risks
The report comes amid increasing warnings from AI researchers and industry insiders over the possibility that advanced AI could pose serious risks to humanity.A senior Anthropic safety researcher recently warned that there was a greater than 10 per cent chance that AI could “kill all humans” within the next decade. OpenAI chief scientist Jakub Pachocki has also called for “voluntary slowdowns” in AI development until adequate safeguards are in place.US Democratic senator Bernie Sanders has meanwhile introduced legislation seeking to temporarily pause advanced AI development and ban artificial superintelligence.US President Donald Trump, however, has dismissed concerns that AI could lead to human extinction. Speaking to reporters on Thursday, Trump said he was worried that the US could be in a “very bad position” if it failed to lead the global AI race.Asked whether he was concerned that AI could lead to human extinction, Trump said: “No, I don’t have any.”Anthropic’s report adds to a rising body of evidence that increasingly powerful AI systems can be used not only for legitimate scientific and technological work but also for activities that could pose security and public safety risks.