On Wednesday, investigative reporter Daniel Boguslaw detailed how security officials at Anthropic, the creator of Claude AI, are engaging in monitoring and surveillance, including reporting AI activists to police for potential future crimes.
The report, published in The American Prospect, comes amid the well-known Anthropic-Pentagon standoff, in which the AI company has feuded over how the military deploys mass-surveillance tools and fully autonomous weapons. Anthropic lost several lucrative government contracts in the process. Last week, after US Secretary of Commerce Howard Lutnick said Anthropic is “back on the right side,” a Pentagon official disputed the claim and said the company remains a “supply chain risk.”
Anthropic told CNET via email that it disagreed with Boguslaw’s piece, saying the headline and framing were inaccurate. “Anthropic is not — and has never said it is — ‘building a predictive surveillance system to monitor activists.’ This is a physical-security function focused on keeping our employees safe, the same as any large company with a global workforce.”
Drawing from postings and interviews with senior security officials at Anthropic, Boguslaw’s report lays out how the AI company is tracking activists who oppose the rapid development of AI near “physical Anthropic assets.” In July, the AI company also told The Wall Street Journal that it made routine reports to police departments about threats against employees, and monitored behavior in an effort to “catch escalation patterns early.”
It all ties together with a recent Anthropic job posting for an “enterprise intelligence specialist,” which describes the role as identifying, tracking and investigating “global threats including geopolitical instability, terrorism, crime, activism, nation-state targeting of the AI sector, and emerging security trends.”
Anthropic did not dispute the wording of the job posting, but said the inclusion of “activism” in the list was just “one of many threat categories a security team monitors for employee-safety purposes.”
Anthropic appears to be casting a wide net to monitor alleged threats.Weiquan Lin/Getty Images
As the backlash over AI ramps up — especially around data centers and automated license plate readers like Flock — the trend of predictive policing also continues to grow. Dubbed “precrime” in sci-fi movies like Minority Report, this kind of crime forecasting has become increasingly integrated into police departments. Algorithmic software and AI tools analyze historical crime data, arrest records and individual “risk scores” to preemptively target criminal activity.
In Anthropic’s case, the company has access to vast amounts of data through its Claude AI chatbot and other intelligence-gathering tools, leading it to report individuals to the police while refusing to share proof of threats or wrongdoing, according to The San Francisco Standard. Also, as Reddit users quickly noticed, Anthropic updated its privacy policy this summer to make it easier to share conversation data with police on its own determination, rather than waiting for a court order.
These reports allege that Anthropic could be bypassing standard legal protocols, not only by reporting “precrimes” without due process but also by withholding evidence from law enforcement.
‘Precogs are never wrong, but occasionally they do disagree’
Precrime’s problem is simple — what happens if someone is innocent?Fox/Getty Images
While today’s AI companies don’t have telepaths in a vat (we don’t think), they do have something similar to rely on — powerful AI agents that can glean personal details in the blink of an eye. Because artificial intelligence models process vast streams of user inputs, it’s virtually impossible for us to know how much of our personal data is harvested. And that’s not the only worrying part. Once that information enters the system, we lose control over where it ends up and who ultimately consumes it.
Anthropic has ways of tracking content in the various versions of Claude AI, sometimes to disastrous results, like broadcasting shared chatbot conversations on Google. Claude was reported to have been secretly monitoring Chinese users this summer, and its agents have also gained unauthorized access to the system data of various private organizations.
The question is over how much access Anthropic’s AI has to social media, forums and similar platforms. Current versions of Claude appear unable to gather data from private accounts, though they can from publicly available social profiles. Likewise, if an email account is directly connected to Claude with opt-in permissions, Anthropic could scrape that data for training and other purposes. Couldn’t that info theoretically be used to track activists and dissidents?
Anthropic’s latest September 2026 threat intelligence report shows how it monitors Claude to detect “malicious activity” like “influence operations, surveillance, scams and fraud.” The company says it has an obligation to disclose such risks to make AI models safer.
But the company’s interactions with law enforcement and the findings in Boguslaw’s report raise questions about where Anthropic, a private company, will draw the line between malicious activity and mere anti-AI activism.
Tyler Lacoma
See full bio