A new threat report from AI developer Anthropic has identified specific ways its Claude model could be manipulated to facilitate harmful activities, including the development of kamikaze drones, the tracking of political dissidents, and the execution of dangerous biological research. The report serves as a practical assessment of current AI risks, shifting the conversation from hypothetical long-term existential threats to immediate security concerns.

According to the findings, the primary danger lies in the ability of AI to act as a force multiplier. Anthropic notes that tasks which previously required large teams of specialized engineers, hackers, or intelligence officers can now be performed by a small group of individuals utilizing AI tools. By lowering these technical barriers, the technology potentially empowers state and non-state actors to conduct operations that were previously beyond their reach.

While the report focuses on the potential for exploitation, it also underscores the ongoing challenge for AI labs to balance model utility with safety guardrails. The document emphasizes that as models become more capable, the risk of them being repurposed for espionage, repression, or weaponization increases. Anthropic’s analysis suggests that the current landscape of AI development is already being navigated by adversaries looking to leverage these systems for strategic advantages in warfare and domestic control.