Anthropic is building an internal surveillance and threat-prediction operation that monitors activists, according to an investigation published by The American Prospect on 9 September 2026. Reporter Daniel Boguslaw documents a corporate intelligence function that tracks protest activity around company executives and facilities, contracts with an outside risk-detection vendor for real-time alerts, and advertises for staff to investigate "activism" in the same breath as terrorism and nation-state threats.
Key facts
- The headline number: an Anthropic job posting for an Enterprise Intelligence Specialist paying $180,000–$230,000, tasked with tracking "activism" alongside terrorism and geopolitical instability.
- When: published 9 September 2026; the job posting went up the previous month.
- Who: reported by Daniel Boguslaw for The American Prospect.
- Primary source: The American Prospect investigation.
The most concrete piece of evidence is not the job ad but a podcast appearance. Keon Ellison, who manages Anthropic's Global Security Operations Center, described handling a planned demonstration in operational detail. "Samdesk gave us about 60 minutes of advanced notice that the protest organizers had moved the timeline," he said. The company used that window to reroute an executive through a service entrance.
Samdesk is a commercial risk-detection service that scans public signals for emerging incidents and pushes alerts to corporate clients. Plenty of large companies buy this kind of feed. What the Prospect argues is distinctive is the combination: a live alerting pipeline, a stated interest in anticipating events rather than reacting to them, and a hiring document that files peaceful protest under the same heading as organised violence.
A useful way to think about the mechanism is a weather alert service repurposed for people. Samdesk-style tools ingest a wide spray of public chatter and flag anomalies — the way a storm tracker flags a pressure drop. Applied to hurricanes, an hour of warning is straightforwardly good. Applied to a protest, the same hour of warning becomes an hour in which a company knows where demonstrators will be before they arrive, and can act on it. The technology does not change; the thing being forecast does.
The reason this is a story about Anthropic specifically, rather than corporate security generally, is that Anthropic has spent years positioning itself as the lab that declines uses other labs would take. Its usage policy places restrictions on surveillance applications, and the company has publicly leaned on that stance. An internal programme that catalogues activism as a threat category sits awkwardly beside that public posture — not necessarily in contradiction of it, since a company securing its own staff is not the same as selling surveillance software, but close enough that the question is fair.
It also lands on an unusually bad day for the company. Hours earlier, a pretraining researcher had resigned saying Anthropic understands the stakes of the AI race but is losing to it, and the company's own alignment lead had publicly agreed that there is no plan for superintelligence. Two unrelated stories, one Tuesday.
The honest caveat is the load-bearing one, and it is worth stating plainly: there is a real difference between is building and is hiring for, and between monitoring protests near our offices and reporting people to police before they commit crimes. The Prospect's framing is the stronger of the two readings. What the documents directly support is a job posting, a vendor relationship, and one manager describing one rerouted executive. The "pre-crime" characterisation — repeated in follow-up coverage from Common Dreams — is the reporter's interpretation of where that architecture points, not a quoted description of a shipped system. Readers should hold those apart.
Anthropic did not respond to the Prospect's request for comment. The company's careers page is the place to check whether the posting remains live; postings of this kind are often pulled once they attract attention.
Why it matters beyond one company: AI labs are becoming, very quickly, some of the most heavily protested institutions in the technology industry, and they are also the institutions with the deepest capability to process signals about people at scale. The gap between "we bought an alerting service like any large employer" and "we built a system to anticipate our critics" is where the next several years of this argument will be fought. The context is not encouraging — the story climbed Hacker News the same day a New Yorker piece on the licence-plate surveillance firm Flock was sitting near the top of the same board, and Ground Truth has previously covered an American city letting its Flock contract lapse. Readers are primed for this shape of story, which is a reason to be more careful with it, not less.
Originally published on Ground Truth, where every claim is checked against the primary source.
Top comments (0)