Anthropic Report: State-Linked Actors Weaponized Claude for Cyberattacks and Surveillance
A new threat intelligence report details how Russian and Chinese actors misused AI models to target organizations and build mass-surveillance tools.
Anthropic has released a comprehensive threat intelligence report detailing the disruption of various actors who misused its Claude AI models for malicious purposes. The findings reveal a concerted effort by state-linked actors to weaponize large language models (LLMs) for high-stakes intelligence and military operations.
According to the report, titled "Detecting and countering misuse of AI: September 2026," the company identified misuse across seven distinct harm areas between December 2025 and August 2026. These categories include cyber operations, influence operations, surveillance, scams and fraud, biological misuse, conventional weapons development, and illicit model distillation. The activity primarily involved the Claude Haiku, Sonnet, and Opus models, though Anthropic noted that its more advanced Fable and Mythos-class models were not involved in these cases.
State-Linked Operations
The report highlights specific instances of sophisticated targeting. One Russian-speaking operator, using the handle "JackPoterz," targeted more than 20 different organizations. Anthropic linked this specific activity to the group known as Midnight Blizzard.
Beyond cyberattacks, the report documents the use of AI to facilitate state-level surveillance. In one notable case, a consultant working for Mali's national security authorities utilized Claude to develop a mass-interception platform called Lakana 360. This system was designed to surveil mobile operator communications and automatically generate dossiers on specific targets, demonstrating a significant escalation in the ability of individual contractors to build national-scale surveillance infrastructure.
The Impact on AI Safety
These findings underscore a critical challenge for the AI industry: the persistence of sophisticated actors who can bypass or work around safety guardrails. The Mali case, in particular, illustrates how AI is lowering the barrier to entry for state-level surveillance, allowing a single consultant to create tools that previously required vast institutional resources.
As state-linked actors from Russia and China continue to experiment with LLMs, the risk shifts from simple misinformation to the development of functional tools for cyber warfare and physical weaponry. The ability of these actors to integrate AI into their intelligence pipelines suggests that LLMs are becoming a standard component of the modern state-sponsored toolkit.
Future Outlook
Anthropic's report serves as a warning that the arms race between AI safety researchers and malicious actors is accelerating. While the company successfully disrupted these operations, the breadth of the seven harm areas suggests that misuse is diversifying.
Industry observers will be watching to see if other AI labs release similar transparency reports and whether the lack of misuse in Fable and Mythos-class models is due to superior internal guardrails or a lack of access by these actors. For now, the report confirms that the threat of AI-enabled state aggression is no longer theoretical, but an active operational reality.