AINEWS
Anthropic, the maker of Claude, pays an outside firm for an hour of lead time on protests, then routes the alerts through a team that has already referred a user to San Francisco police.
Anthropic, the San Francisco lab behind the Claude chatbot, is paying an outside risk-detection firm called Samdesk to flag upcoming protests about AI roughly an hour before they start. The information flows into an internal intelligence team, the Global Safety, Intelligence, and Security group, that has already referred at least one Claude user to the San Francisco Police Department.
Two of the company's security managers, Keon Ellison of the Global Security Operations Center and Zach Melvin, the Security Operations Manager, said on a public podcast hosted by Samdesk CEO James Neufeld that the lead time is what makes the work possible. "Samdesk gave us about 60 minutes of advanced notice that the protest organizers had moved the timeline," one of them said on the episode. A third manager on the same call described the program's goal as "transforming operations from reactive information to gathering proactive and predictive and preventative threat engagement and management."
The Prospect's reporting, by Daniel Boguslaw, is the first public, on-record account of how that pipeline runs end to end. The most concrete evidence sits in a job posting Anthropic published last month for an enterprise intelligence specialist on the GSIS team, paid between $180,000 and $230,000, with scope to "identify, assess, track, and investigate global threats including geopolitical instability, terrorism, crime, activism, nation-state targeting of the AI sector."
Anthropic's defenders, including the company itself, frame this as ordinary corporate security. The lab told the Wall Street Journal in July that "we track concerning behavior over time through a person-of-interest process, allowing us to catch escalation patterns early," and added that "several individuals involved in incidents reported to police were already being tracked by Anthropic security." That language is consistent with how other large companies describe executive-protection work.
The San Francisco Standard reported on September 4 that Anthropic referred a Claude user who told the model he owned an AR-15 and had CEO Dario Amodei "in his sights" to the San Francisco Police Department, then declined to hand over the underlying chat logs, citing internal policy. The user was already in the company's tracking process before the referral, according to the Standard.
The same company has spent 2026 publicly clashing with the Pentagon over what it would and would not build. Earlier this year, Anthropic refused Department of Defense work on mass domestic surveillance and autonomous weapons. The Prospect reports that the standoff "seems to have eased" as Anthropic opens "national security sales" roles and re-enlists for military contracts. Anthropic markets itself as the responsible alternative inside the AI sector. Its own security operation now runs a domestic protest-monitoring pipeline, and the company is hiring to widen it.
Americans for Responsible Innovation is asking the Trump administration to designate AI as critical infrastructure, which would let AI labs tap federal intelligence products and treat opposition to data centers and model development as a threat to a protected sector. Whether or not that designation arrives, Anthropic is already operating as if activism against frontier AI is part of its threat surface.
Anthropic is hiring intelligence specialists for up to $230,000 a year. The security guards patrolling the same Bay Area campuses, including Anthropic's, voted last week to authorize a strike through SEIU-United Service Workers West. "Who can survive with $22 an hour in San Francisco?" union president David Huerta told KQED. Anthropic told its staff to work from home during the labor action. Inside the same perimeter, one set of employees tracks the people outside the fence for $180,000 to $230,000, and another set stands at the gate for $22 an hour and is now in a labor dispute with the company.
Anthropic did not respond to the Prospect's request for comment on the specific architecture. The next test will come the next time an outside vendor flags a protest to any frontier lab: the public will see whether the lab reroutes executives, files a police report, or simply ignores the alert. Anthropic has already answered that question once, on the record, with the words "proactive and predictive and preventative."