AI Safety: Claude Misused in Biological Research

22:29 / 11.09.2026·30·Technology

The rapid development of AI technologies brings not only new opportunities but also serious security threats. In its latest report, Anthropic, the creator of the popular Claude AI, revealed several dangerous cases where AI was potentially used in research aimed at creating biological weapons. This is reported by Ixbt.com reports.

According to ixbt.com, these incidents were recorded between December 2025 and August 2026. A key problem was the difficulty in determining the true intent of research during the switching process, as the same scientific knowledge can be used both to create beneficial medicines and to develop highly dangerous pathogens.

Dangerous biological research and covert attempts

In one of the identified cases, researchers were working on the mosquito-borne chikungunya virus. Claude's safety system blocked a request aimed at preparing a grant application for enhancing virus functions. After it was discovered that a military research institution was behind the request, Anthropic restricted the service, but the scientists continued their work through other channels.

Analysis showed that users employed various tools to bypass models with stronger safety mechanisms. Specifically:

An intermediary platform was identified that redirected rejected biological requests to other AI models with weaker safety restrictions.
In a study on highly pathogenic avian influenza, scientists used Claude to plan experiments for several weeks and continued the work using lower-power versions.

It was also reported that three other cases involved orthopoxviruses, various toxic compounds, and toxins.

Cybersecurity and Midnight Blizzard activity

Anthropic's report states that AI has been misused not only in biological fields but also in cybercrime. The company linked one of the identified cyber operations to the Midnight Blizzard group, which is considered to be associated with Russian hackers.

As part of this campaign, AI was used to automate phishing attacks, prepare network infrastructure, create malware, and steal data. Such actions targeted more than 20 organizations belonging to government, diplomatic, and defense structures in Ukraine and European countries.

Anthropic experts emphasize that as the scientific analysis capabilities of AI expand, the potential for misuse also increases. The company has blocked high-risk accounts, provided data to relevant authorities, and further tightened restrictions on high-risk biological queries.