Public Policy & Governance

U.S. District Judge Rita Lin granted Anthropic a preliminary injunction pausing the Department of Defense's designation of the company as a supply chain risk, following Anthropic's refusal to permit military use of its Claude AI model in autonomous weapons systems. The temporary order, issued by the Northern District of California, prevents enforcement of punitive measures while the court considers the company's substantive legal challenge. The Guardian reports the designation stemmed from Anthropic's explicit prohibition on defense department applications involving lethal autonomous systems. However, Politico notes the company must still secure a separate stay from Trump-appointed judges in the D.C. Circuit Court of Appeals, with legal experts cautioning the California victory may be 'premature' given the dual-track nature of the dispute.

The case represents the first significant judicial examination of federal authority to compel commercial AI companies into defense applications through supply chain risk designations — a mechanism that carries severe business consequences including exclusion from federal contracts and reputational damage. The dispute exposes fundamental tensions between voluntary corporate AI safety positions and government prerogatives to access dual-use technologies for national security purposes. The preliminary injunction provides Anthropic tangible relief to show customers and investors, though Politico reports the company faces an uphill battle convincing appellate courts to limit executive branch discretion over defense procurement.

Senator Bernie Sanders and Representative Alexandria Ocasio-Cortez introduced federal legislation to impose a moratorium on new AI datacenter construction, arguing the pause would provide time to establish 'strong, federal guardrails' for AI development amid what they characterize as an unprecedented energy crisis. The Guardian reports the lawmakers criticized the absence of serious congressional discussion about AI's societal impact despite its significance, framing the infrastructure buildout as outpacing democratic deliberation and environmental sustainability constraints.

The proposal represents the first congressional attempt to directly constrain AI infrastructure development through construction restrictions rather than operational regulations. The bill's viability remains uncertain given Republican control of both chambers and industry opposition, but it signals a strategic shift among progressive Democrats toward treating physical infrastructure as a regulatory leverage point. The timing coincides with mounting evidence of datacenter energy demands straining regional grids and conflicts with renewable energy targets, though the legislation does not specify duration of the proposed moratorium or criteria for lifting it.

The Center for Democracy & Technology submitted formal comments on NIST's draft guidance for automated benchmark evaluations of language models (AI 800-2), while the Electronic Frontier Foundation filed a FOIA lawsuit against the Centers for Medicare & Medicaid Services demanding disclosure of records about a multi-state AI program evaluating medical care requests affecting millions of seniors. CDT engaged with NIST's Center for AI Standards and Innovation on technical evaluation practices, though the organization did not disclose specific concerns in its public announcement. EFF argues the CMS program 'tasking an algorithm with making determinations about treatment can create unwarranted—and even discriminatory—delays or denials of necessary medical care,' yet CMS has released minimal information about the system's design, vendor, or performance metrics.

The dual civil society actions reflect growing frustration with opacity in both AI standards-setting and operational deployment in high-stakes government applications. The CMS case is particularly significant because Medicare AI deployment represents direct federal use affecting vulnerable populations with minimal public oversight or accountability mechanisms. EFF's resort to litigation indicates CMS has declined voluntary disclosure despite the program's scale and impact. The NIST comment period, while more routine, shows civil society organizations attempting to shape technical evaluation standards before they become embedded in procurement requirements or regulatory frameworks.

CDT Europe reports March brought significant movement on the AI Omnibus package — technical amendments to the AI Act addressing implementation gaps — with trilogue negotiations advancing between the European Parliament, Council, and Commission. The bulletin notes policymakers are also developing positions on copyright frameworks for generative AI and new rules governing AI-generated content labeling. Multiple consultations opened during March, indicating the EU is moving from the AI Act's passage into detailed rule-making across copyright, content authenticity, and sectoral applications.

The AI Omnibus represents the EU's first major technical correction cycle for its landmark AI Act, addressing ambiguities and implementation challenges that emerged after the legislation's adoption. Copyright consultations are particularly consequential given unresolved tensions between rights holders and AI developers over training data, with potential implications for EU-U.S. regulatory divergence. The content labeling initiatives respond to concerns about synthetic media in electoral contexts and public discourse, though enforcement mechanisms remain undefined.