Anthropic has sounded a fresh alarm over the growing misuse of advanced artificial intelligence, revealing that its Claude models have been targeted by state-linked groups, criminals and other organisations for activities involving biological research, military applications, surveillance, cyber espionage and influence operations.
In a detailed threat intelligence report covering incidents identified between December 2025 and August 2026, Anthropic said it disrupted multiple cases that it described as among the most notable and sophisticated misuse activity it has detected so far. The cases involved suspected actors linked to China, Russia and Iran, as well as financially motivated criminals, spyware operators and politically driven groups.
The findings highlight a growing challenge for frontier AI companies: as models become more capable, they can potentially accelerate complex activities that previously required teams of specialised researchers, engineers and analysts.
Bioweapons Research Raises Major AI Safety Concerns
Anthropic identified biological research as one of its most sensitive areas of concern. The company documented five cases involving users seeking assistance with research that could potentially contribute to harmful biological applications.
Anthropic noted that these cases are particularly difficult to assess because legitimate scientific research and dangerous dual-use research can sometimes involve similar technical questions.
“It's an incredibly nuanced situation,” Jacob Klein, Anthropic’s head of threat intelligence, said while discussing the challenge.
The company said some users attempted to bypass its safety controls after Claude refused certain requests. Anthropic responded by blocking accounts, applying stronger refusals, restricting access to less capable models and conducting proactive investigations for suspicious activity.
AI Moves Into Military Applications
Anthropic also reported multiple attempts to use Claude in military and weapons-related work. The company said a China-based actor used Claude across several areas connected to an anti-torpedo weapons programme and assessed that the individual was associated with China's defence industry. The model was reportedly used for technical documentation and analysis of publicly available information concerning military systems.
Anthropic also identified cases linked to Russia and Yemen involving attempts to apply AI to advanced military technologies. The disclosures demonstrate how AI systems could increasingly become part of defence-related research and development, particularly by helping users process technical information, prepare documentation and automate specialised tasks.
Surveillance and Espionage Add to the Risk
Weapons development was only one part of Anthropic's findings. The company identified nine surveillance-related cases involving alleged state-linked activity. According to Anthropic, China-linked operations used Claude to analyse information and create profiles involving ethnic, religious and political groups. Iran-linked activity similarly involved attempts to profile individuals and analyse large volumes of social-media information.
The company also identified cyber espionage operations in which Claude was allegedly used to support reconnaissance, coding and other activities targeting government, defence and diplomatic organisations. Anthropic said it banned accounts associated with the activity and strengthened its detection systems to identify behavioural patterns rather than relying solely on individual prompts.
AI Companies Face Model-Distillation Threat
Anthropic's report also highlights another emerging concern: competing AI developers attempting to extract capabilities from established models. The company said it identified large-scale efforts involving several Chinese AI companies, including Alibaba, Moonshot AI and DeepSeek, which it described as illicit model distillation.
According to Anthropic, one campaign involved thousands of accounts and generated more than 150 million exchanges with Claude over several months. The company said such activity can allow competitors to obtain large volumes of model-generated responses that may help improve their own systems.
Anthropic has since introduced additional identity verification, strengthened extraction detection and taken action against accounts associated with suspected abuse.
The AI Safety Challenge Is Getting Bigger
Anthropic's latest report illustrates how the AI threat landscape is changing. Misuse is increasingly moving beyond individual bad actors and isolated harmful prompts towards organised operations involving governments, criminal networks and technology companies. The company said it is sharing relevant intelligence with authorities, researchers, industry partners and affected organisations to improve collective defences.
As frontier AI models become more powerful, the challenge for developers will increasingly be about identifying not only what users ask AI systems to do, but who is behind the activity, how the technology is being deployed and whether multiple seemingly harmless interactions form part of a much larger operation.
Anthropic's findings suggest that AI safety is entering a new phase—one where protecting powerful models may require continuous monitoring, stronger safeguards and cooperation across the entire technology and security ecosystem.
𝐒𝐭𝐚𝐲 𝐢𝐧𝐟𝐨𝐫𝐦𝐞𝐝 𝐰𝐢𝐭𝐡 𝐨𝐮𝐫 𝐥𝐚𝐭𝐞𝐬𝐭 𝐮𝐩𝐝𝐚𝐭𝐞𝐬 𝐛𝐲 𝐣𝐨𝐢𝐧𝐢𝐧𝐠 𝐭𝐡𝐞 WhatsApp Channel now! 👈📲
𝑭𝒐𝒍𝒍𝒐𝒘 𝑶𝒖𝒓 𝑺𝒐𝒄𝒊𝒂𝒍 𝑴𝒆𝒅𝒊𝒂 𝑷𝒂𝒈𝒆𝐬 👉 Facebook, LinkedIn, Twitter, Instagram