Anthropic builds predictive surveillance system to track activists and protests
Anthropic is building a predictive surveillance system to track activists and monitor protests near its offices, according to an investigation by The American Prospect. The company also reported a user to police for statements made to its Claude chatbot without providing law enforcement with supporting evidence of wrongdoing.
What happened
According to an investigation by The American Prospect, Anthropic is building a predictive surveillance system to track and monitor activists opposed to artificial intelligence development. The company issued a job posting seeking an intelligence specialist to 'identify, assess, track, and investigate global threats including geopolitical instability, terrorism, crime, activism, nation-state targeting of the AI sector,' using a 'pre-crime' approach to predict incidents before they occur.
Anthropic security officials described using risk detection company Samdesk to monitor protests near company offices and provide advance notice of planned demonstrations. In one incident, a security manager stated that after learning through Samdesk that protest organizers had moved their timeline, 'that extra hour was critical' to prevent Anthropic executives from encountering the disruption.
Anthropic reported a San Francisco man to police after he told the Claude chatbot that he had purchased a rifle and had CEO Dario Amodei 'in his sights.' The company reported the conversation to authorities but refused to provide law enforcement with the chat transcripts, effectively reporting the man without sharing tangible evidence of misconduct.
Reported impact
Civil society and activists opposed to AI development face targeted surveillance based on their speech and perceived threat level, raising concerns about privacy and freedom of association.
Index analysis
A legal expert noted that as AI companies 'become more hated, there's more pressure on them to over-disclose knowing that some of the people they identify for law enforcement shouldn't be targeted.' The approach of reporting user conversations to police without providing supporting evidence raises questions about due process and the reliability of threat assessments.
Governance controls
Anthropic's security operations employ a stated goal of 'transforming operations from reactive information to gathering proactive and predictive and preventative threat engagement and management.' However, the company's practice of reporting users to police without providing chat transcripts demonstrates a gap between stated security principles and actual implementation.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.