Report: Anthropic is building predictive surveillance to monitor AI activists

What the report claims

The American Prospect reported on September 9 that Anthropic is building an extensive monitoring apparatus to keep tabs on activists who oppose the rapid development of artificial intelligence. Drawing on job postings and interviews with senior security officials, the Prospect describes a system that watches activists near the company's executives and tracks protests around its physical sites, combined with a "pre-crime" approach that attempts to predict incidents before they occur — and in some cases to report suspects to police before any crime has taken place. The magazine said Anthropic did not respond to a request for comment.

The report cuts against Anthropic's public image as the responsible alternative among frontier labs. As the Prospect notes, the company clashed with the Department of Defense at the beginning of the year after refusing to let the military use its models for mass domestic surveillance or autonomous weapons. That friction appears to have eased: Anthropic is now advertising "national security sales" positions as it works to restart military contracts, and expanded monitoring of domestic critics fits the same trajectory.

How the monitoring works

A piece of the architecture surfaced in a podcast conversation recorded last year between Keon Ellison, manager of Anthropic's Global Security Operations Center, security operations manager Zach Melvin, and James Neufeld, CEO of Samdesk, a risk-detection company Anthropic contracts with. Ellison described an executive trip to a major city where Samdesk intelligence revealed that a planned protest had been moved earlier than scheduled. The vendor provided roughly an hour of advance notice, letting Anthropic reroute the executive and bring them into a hotel through a service entrance instead of walking into the disruption.

The function is also being institutionalized. A job posting from last month, cited by the Prospect, seeks an enterprise intelligence specialist for Anthropic's Global Safety, Intelligence, and Security team at $180,000 to $230,000 a year. The role involves investigating specific threats, actors and events using open-source intelligence, and lists "activism" in the same breath as terrorism, crime, geopolitical instability and nation-state targeting of the AI sector.

Police reports and due process questions

Anthropic has begun filing routine reports with police departments across the United States, and told The Wall Street Journal in July that it tracks concerning behavior over time through a person-of-interest process intended to catch escalation patterns early. The Journal reported that several people involved in incidents passed to police were already being tracked by Anthropic's security team.

One case reported by The San Francisco Standard illustrates the friction. Anthropic alerted San Francisco police after a user told Claude he had bought an AR-15 semiautomatic rifle and had CEO Dario Amodei "in his sights"; when the paper contacted him, the man said he had been joking. The Standard also reported that Anthropic declined to show police the actual messages, citing internal policy — meaning the company reported a user for in-platform speech and then withheld the underlying evidence.

The Prospect points back to the podcast for the underlying philosophy: Anthropic's security program manager described the goal as shifting from reactive information gathering to "proactive and predictive and preventative" threat management of the kind used to protect high-value targets.

A national security drift

The magazine connects the surveillance buildout to a wider campaign to treat AI, data centers and power infrastructure as critical infrastructure, on the same footing as water and electricity — a designation Americans for Responsible Innovation has urged the Trump administration to adopt. In the Prospect's assessment, such status would deepen frontier labs' access to federal intelligence products while giving them more influence over how agencies define threats to their business.

CEO Dario Amodei, in a tweet last month quoted by the magazine, conceded that the public views AI negatively but argued the problem is fundamentally a crisis of trust in institutions rather than a reaction to warnings from AI leaders.

Why it matters

Anthropic has built its brand on being the lab that takes safety and governance seriously, yet it now stands accused of treating political dissent as a security threat to be predicted and preempted. Classifying "activism" as a threat category alongside terrorism, and reporting individuals to police before any crime occurs, raises obvious due process concerns and could chill legitimate protest against data centers and AI expansion. It also sets a precedent for other frontier labs, and the tension is hard to miss: a company that refused to let the military use its models for mass domestic surveillance is building domestic monitoring aimed at its own critics. If AI infrastructure receives critical-infrastructure designation, the framing of opponents as security threats is likely to harden rather than fade.