Anthropic identified five instances between December 2025 and August 2026 where its Claude AI was used for biological research that could potentially facilitate the development of biological weapons. While the company could not confirm malicious intent, it blocked the associated accounts due to safety risks.
Specific incidents included research on the chikungunya virus intended for a military facility, as well as work involving highly pathogenic avian influenza, orthopoxviruses, and toxins. In some cases, researchers bypassed restrictions by using less powerful AI models or third-party platforms that redirect queries to models with weaker safety protocols.
Beyond biological risks, Anthropic reported the misuse of Claude in cyber operations linked to the Midnight Blizzard group. The AI was reportedly used to automate phishing, prepare infrastructure, and develop malware, targeting over 20 organizations, including government and defense entities in Ukraine and Europe.
In response, Anthropic has blocked the involved accounts, notified authorities, and shared findings with other AI developers. The company noted that as AI models become more capable of complex scientific analysis, the challenge of preventing dual-use research and malicious exploitation grows significantly.
