Now loading...
Anthropic launched a legal challenge against the Department of Defense on Monday, contesting its classification as a supply chain vulnerability. In a swift show of solidarity, almost 40 staff members from OpenAI and Google, among them Jeff Dean, Google’s top scientist and head of the Gemini project, submitted an amicus brief backing Anthropic’s position just hours afterward. The filing underscores worries about the Trump administration’s ruling and the broader dangers tied to advanced AI applications.
This development caps a tense period for Anthropic, coming after the Trump administration marked the firm as a national security hazard, a label usually applied to overseas entities posing threats. The move followed Anthropic’s refusal to budge on two firm boundaries for military applications of its AI: widespread surveillance within the United States and weapons systems operating independently without human input. Talks collapsed, leading to heated public exchanges, while competitors like other AI outfits inked deals permitting broad legal applications of their tech.
Beyond barring Anthropic from defense deals, the risk label extends to any firm incorporating its tools in Pentagon-related projects, compelling them to abandon Claude to keep their agreements intact. Yet as the initial AI approved for handling sensitive intelligence, Anthropic’s software is woven into military operations. Notably, shortly after Defense Secretary Pete Hegseth revealed the designation, reports emerged of U.S. forces deploying Claude in an operation that eliminated Iran’s supreme leader, Ayatollah Ali Khamenei.
The amicus brief contends that labeling Anthropic a supply chain risk amounts to unjust reprisal damaging broader societal benefits. It stresses the legitimacy of Anthropic’s boundaries, arguing that AI-driven domestic surveillance threatens core democratic principles, regardless of oversight, and that self-governing lethal systems demand urgent safeguards. The document urges reevaluation of these limits to address emerging perils.
Signatories identify as engineers, researchers, scientists, and experts at leading American AI development centers. They note their hands-on experience with powerful models used in critical areas like security, policing, and warfare. Speaking individually, not for their employers, they highlight the urgency of aligning AI deployment with existing laws and ethics to avert misuse.
Regarding surveillance, the experts explain that while vast data from cameras, tracking, online activity, and finances already surrounds citizens, AI could stitch it all together into a seamless monitoring network. Currently isolated, these sources might merge via AI to link identities, movements, purchases, connections, and habits across populations in real time, amplifying privacy erosion.
On autonomous weapons, the group points out their limitations in unfamiliar scenarios, where they falter in target identification and fail to weigh unintended consequences as humans do. Prone to errors like fabricating details, these systems necessitate human judgment before any deadly action, particularly since their decision paths remain opaque to both users and creators.
Despite varied viewpoints, the contributors agree that current top-tier AI carries significant hazards in unchecked surveillance or lethal autonomy, calling for measures like built-in limits or policy constraints to mitigate them.
