Anthropic sues, claiming government overreach threatens AI safety decisions

Anthropic’s complaint, filed in the U.S. District Court for the Northern District of California, accuses 17 federal agencies and the Executive Office of the President of abusing statutory authority to coerce the company into abandoning two core safety guardrails on its Claude model—namely, a prohibition on lethal autonomous weapons and a ban on mass surveillance of U.S. citizens (The‑Decoder). The suit argues that the Department of Defense simultaneously threatened to invoke the Defense Production Act (DPA) to commandeer Claude for military use while also moving to blacklist the company as a “security risk,” a contradictory stance that, according to the filing, violates the principle that a product cannot be deemed both essential and dangerous under the same legal framework (The‑Decoder).

At the heart of Anthropic’s legal challenge is the interpretation of 10 U.S.C. § 3252, a statute originally crafted to address foreign adversary sabotage of information systems. The complaint contends that the government’s reliance on this provision is misplaced because the law’s definition of “foreign adversary” is limited to nations such as China, Russia, Iran, North Korea, Cuba and Venezuela, none of which are directly implicated in the agency’s actions against Anthropic (The‑Decoder). By stretching the statute to cover domestic policy decisions about AI safety, the government, the filing alleges, is exceeding its jurisdiction and setting a precedent that could allow regulators to punish companies for ethical constraints that align with broader public‑interest concerns.

Anthropic’s lawsuit also highlights the broader regulatory climate that has emerged since the passage of the AI Risk Management Framework and subsequent executive orders. While many AI firms have opted to quietly adjust their product roadmaps to avoid confrontation, Anthropic’s refusal to strip away its safety layers has placed it at the center of a policy showdown. The company’s stance underscores a growing tension between the federal push for rapid AI deployment in defense and intelligence contexts and the industry’s self‑imposed safeguards designed to prevent misuse (The‑Decoder). By challenging the DPA threat, Anthropic seeks a judicial clarification that could limit the government’s ability to weaponize AI without explicit congressional authorization.

Legal scholars cited in the filing warn that a ruling in Anthropic’s favor could reshape the balance of power between private AI developers and federal agencies. If the court determines that the DPA cannot be wielded to force compliance with safety‑related policy demands, it would curtail the executive branch’s leverage over emerging technologies and reinforce the notion that safety guardrails are a permissible exercise of corporate discretion (The‑Decoder). Conversely, a dismissal could embolden regulators to impose stricter controls on AI systems deemed “critical” to national security, potentially forcing other firms to compromise on ethical safeguards to remain in government supply chains.

The case arrives at a moment when the AI sector is grappling with mounting pressure to align its products with defense priorities. Recent announcements, such as Microsoft’s rollout of Copilot features that automate software development, illustrate the industry’s pivot toward government‑backed use cases (VentureBeat). Anthropic’s legal battle therefore serves as a bellwether for how far the government can go in dictating the moral parameters of AI, and whether the courts will uphold the industry’s right to embed safety constraints without punitive repercussions.