Anthropic has built a surveillance apparatus designed to keep tabs on activists opposed to the development of artificial intelligence. This was revealed by an investigation based on job advertisements and interviews with company security managers, according to which the system does not simply observe: it aims to predict accidents before they happen, in some cases reporting suspicions to the police even before a crime is committed.
Monitoring covers two fronts. On the one hand, the movements of managers, protected by a service that follows nearby demonstrations in real time. On the other hand, the protests organized near the company’s physical locations, traced with an explicitly defined approach “pre-crime” by one of the security program managers.
The strategy clashes with the image that Anthropic has built in recent years, that of a responsible alternative to OpenAI on the ethical issues of AI. At the beginning of 2026, the company had refused the Pentagon the use of its tools for mass surveillance and autonomous weapons. The open positions for “national security sales” roles now indicate a rapprochement with military contracts.
The surveillance architecture designed by Anthropic
Some details emerge from a podcast interview between Keon Ellisonhead of Anthropic’s Global Security Operations Center, colleague Zach Melvin and James Neufeld, CEO of Samdesk, the risk detection contractor. Ellison recounts a concrete case: a protest ahead of schedule due to permit issues, with Samdesk providing 60 minutes notice enough to divert an executive towards a service entrance of the hotel.
Reporting to the police is now a consolidated practice. In a statement to the Wall Street Journal in July, Anthropic explained that it tracks “concerning behavior over time through a person-of-interest process, which helps identify escalation patterns early.” According to the newspaper, several individuals involved in incidents already reported to the police were under observation by company security.
Another controversial case involves a man who was reported to the San Francisco police after he wrote to Claude that he had purchased an AR-15 semi-automatic rifle and to have CEO Dario Amodei “in the crosshairs”. Contacted by the San Francisco Standard, the man said he was joking. Anthropic, despite having reported the episode, refused to show the agents the actual messages, citing internal policies: it reported a user for what he wrote on the platform, then denying the police direct evidence of the alleged crime.
The expansion of corporate intelligence also passes through a team job announcement Global Safety, Intelligence and Securitywhich seeks a specialist paid between $180,000 and $230,000 annually to “identify, assess, track and investigate global threats” that explicitly include geopolitical instability, terrorism, crime and activism.
The context is part of a broader push, supported by groups such as Americans for Responsible Innovation, to have artificial intelligence and the infrastructure that powers it classified as “critical infrastructures” on a par with electricity and water mains. A similar designation would grant AI labs broader access to federal agencies’ intelligence products, as well as the power to influence how Washington defines threats to its business.
The risk, raised by the investigation, is that civic activism against data centers will soon be labeled extremism. Tension with the company’s non-technical staff remains in the background: security guards patrolling the campuses of Anthropic and OpenAI have authorized a strike over the stalemate in salary negotiations, while the company has invited employees to work from home.
