The narrative surrounding the frontier AI labs has long been one of ethical safeguards and the pursuit of alignment. For years, the industry's leading voices spoke of constitutional AI and the necessity of keeping powerful models away from the machinery of state surveillance. However, a quiet but decisive shift is occurring behind the scenes. The tension is no longer just about whether an AI will hallucinate or leak data, but about how these companies are transforming into intelligence hubs that monitor human behavior in real-time to preempt perceived threats.
The Machinery of Predictive Monitoring
Anthropic has moved beyond traditional corporate security, adopting a pre-crime approach to manage its physical and digital perimeter. The core of this strategy is a partnership with Samdesk, a risk intelligence firm specializing in real-time monitoring. This system is designed not to react to incidents after they occur, but to identify signs of unrest and intercept them before they manifest. In one documented instance, the system flagged a sudden change in a protest schedule in a city where an Anthropic executive was visiting. Samdesk provided a notification 60 minutes before the event, allowing the executive to bypass the crowds entirely by entering the hotel through a service entrance. This represents a transition from security as a protective shield to security as a predictive navigation tool.
To institutionalize this capability, Anthropic is aggressively expanding its Global Safety, Information, and Security (GSIS) team. Recent job postings reveal a search for an Enterprise Intelligence Specialist, a role with a salary range between 180,000 and 230,000 dollars. The mandate for this position is expansive, requiring the specialist to identify and track global threats including geopolitical instability, terrorism, and organized crime. Crucially, the role explicitly includes the tracking of activism. The primary tool for this effort is Open Source Intelligence (OSINT), which involves the systematic collection and analysis of public data to generate intelligence reports on potential adversaries or disruptors.
This intelligence apparatus has already intersected with law enforcement. When a user sent a message to Claude stating they had purchased an AR-15 semi-automatic rifle and intended to target CEO Dario Amodei, Anthropic immediately contacted the San Francisco Police Department. However, a significant detail emerged during this interaction: Anthropic refused to provide the actual message logs to the police, citing internal company policies. This creates a paradoxical operational loop where the company uses its platform to trigger state police action while simultaneously withholding the primary evidence required for a standard investigation.
The Pivot to National Security Infrastructure
This shift reveals a fundamental contradiction in Anthropic's brand identity. Earlier this year, the company maintained a public stance of caution, resisting efforts by the U.S. Department of Defense to integrate its tools into large-scale domestic surveillance or autonomous weapons systems. Yet, the internal trajectory has pivoted toward the state. The company is now recruiting personnel specifically for national security sales, signaling a willingness to re-engage with military contracts and government intelligence frameworks.
This evolution is driven by a broader strategic effort to reclassify AI infrastructure as critical infrastructure. Groups such as the Association for Responsible Innovation (ARI) have been lobbying the Trump administration to categorize AI on the same level as water, electricity, and broadband communications. If AI is officially designated as a national critical asset, the relationship between frontier labs and the state changes entirely. Such a designation would grant companies like Anthropic easier access to intelligence products generated by federal law enforcement and intelligence agencies, effectively merging corporate security with national security.
What is actually changing here is the definition of a threat. By moving from reactive information gathering to a proactive, predictive, and preventative response, Anthropic is no longer just protecting a business; it is adopting the posture of a state actor. The goal is to protect high-value targets and corporate interests under the umbrella of national security, which in turn allows the company to influence how the government defines threats. The asymmetry of data control is the most potent tool in this new arsenal. Anthropic possesses the power to define a user as a threat and mobilize the police, but it retains absolute control over the data that justifies that definition, shielding its internal processes from the very legal scrutiny it invokes upon others.
This obsession with external predictive threats stands in stark contrast to the company's internal realities. While Anthropic invests hundreds of thousands of dollars in OSINT specialists to predict the movements of activists, it has struggled to manage the tangible grievances of its own workforce. Security guards tasked with protecting the company's physical campuses recently declared a strike following a breakdown in wage negotiations. The irony is sharp: the company has built a sophisticated system to predict and avoid a protest on a city street, yet it failed to predict or prevent a labor uprising within its own walls.
The trajectory of the AI industry is moving away from the idealistic pursuit of safety and toward a symbiotic relationship with the security state.
