By Emmanuel Nduka
Artificial intelligence company, Anthropic, has disclosed that it disrupted several attempts to use its Claude AI models for research that could potentially facilitate the development of biological weapons.
The company said the cases were uncovered between December 2025 and August 2026 as part of investigations into the misuse of its artificial intelligence systems.
According to Anthropic, its threat intelligence team identified five separate cases involving biological research that raised concerns because the work could potentially be applied to the development of more dangerous pathogens or toxic substances.
The company said the cases highlighted a growing challenge in controlling advanced AI systems, particularly because the same scientific knowledge that can contribute to medical breakthroughs can also be applied for harmful purposes.
Anthropic said one of the cases involved an attempt to obtain Claude’s assistance in preparing a scientific grant application involving research on the chikungunya virus. The research included work that could alter characteristics of the virus in ways that raised biological safety concerns.
The company said its safeguards blocked the initial request, after which it launched a wider investigation into the activity.
Anthropic subsequently found that the individuals involved had been accessing AI models through an intermediary platform that sought to circumvent regional restrictions and AI safety controls.
The company said it banned the accounts associated with the activity, worked with other technology companies and shared relevant information with government authorities.
Another case involved a researcher using Claude in connection with research into highly pathogenic avian influenza and its adaptation to mammals.
Anthropic said its safety systems prevented the researcher from accessing its more capable models for the sensitive work, limiting the assistance available through Claude.
The company, however, stressed that it was not claiming that the researchers involved necessarily intended to develop biological weapons.
It said the difficulty arose from the inherently dual-use nature of biological science, where research intended to understand disease, develop vaccines or create treatments can also generate knowledge that could potentially be exploited for harmful purposes.
Anthropic said the problem was becoming more serious as AI models grew increasingly capable of supporting sophisticated scientific research.
The company noted that earlier versions of its models had been assessed as being incapable of meaningfully assisting sophisticated users with dangerous biological research.
That assessment, it said, could no longer be made with the same confidence for newer frontier models.
As a result, Anthropic said it had introduced stronger safeguards on its latest models, including restrictions covering a broader range of sensitive biological research.
The company also disclosed cases involving research into toxins and other biological substances, saying some of the activities were connected to state-supported research programmes.
According to Anthropic, the cases demonstrate why simply blocking obviously malicious requests may not be enough to prevent the misuse of increasingly sophisticated AI systems.
The company said sophisticated actors could deliberately disguise their intentions by presenting potentially dangerous research as legitimate scientific work, making it difficult for automated systems to distinguish beneficial research from activity that could create serious security risks.
Anthropic said it had therefore strengthened its monitoring and enforcement mechanisms, while calling for greater cooperation among AI companies, governments and the wider scientific community.
The company warned that biological misuse represents one of the most serious potential risks associated with increasingly capable AI, although it acknowledged that the probability of such scenarios remains uncertain.
It said greater transparency about real-world attempts to misuse AI would be necessary to help policymakers and technology companies understand the emerging risks and develop more effective safeguards.
