Public Policy & Governance

Baltimore's lawsuit against xAI targets the company's marketing of Grok as a general-purpose assistant while allegedly failing to disclose risks of generating nonconsensual sexual content, according to The Guardian. The legal theory relies on consumer protection statutes rather than existing image-based abuse laws, potentially establishing precedent for holding AI model developers liable for foreseeable harmful outputs rather than just their distribution. This represents a shift from targeting individual deepfake creators to pursuing the infrastructure providers, though the case faces significant Section 230 immunity questions that Baltimore will need to overcome.

The timing coincides with increasing reports of AI-generated nonconsensual imagery, but Baltimore's decision to act at the municipal level reflects the vacuum left by federal inaction. The lawsuit may prompt other cities to pursue similar actions, creating a patchwork of local AI liability standards that could pressure Congress to establish federal frameworks — or alternatively, provide ammunition for industry arguments that only federal preemption can prevent regulatory fragmentation.

The December executive order directing federal agencies to sue states attempting AI regulation and withhold funding represents the administration's most aggressive move to establish federal preemption, as reported by The Guardian. The order explicitly supports industry lobbying against state-level constraints, but provides no alternative federal regulatory framework — creating a deliberate void rather than harmonisation. Several states including California and Colorado have existing AI laws potentially affected, though no lawsuits have been publicly filed yet.

The constitutional question centres on whether AI regulation falls under traditional state police powers or whether federal commerce clause authority allows preemption absent specific federal standards. The order's funding threats echo previous failed attempts to coerce states on immigration enforcement, which courts rejected. What makes this particularly notable is the absence of strong legislative champions on either side ahead of midterms — Republicans are divided between libertarian and national security factions, while Democrats lack a coherent position beyond vague calls for 'guardrails'.

Freedom of Information requests reveal the UK government has not conducted any trials involving OpenAI technology eight months after signing a memorandum of understanding promoted as enabling 'AI-led public service reform', reports The Guardian. This follows a pattern where ministerial announcements generate headlines but implementation stalls due to procurement rules, data protection requirements, and departmental capacity constraints. The OpenAI agreement contained no binding commitments or timelines, functioning primarily as a statement of intent.

Meanwhile, Palantir has secured an FCA contract to analyse financial regulation data for fraud detection, according to The Guardian, demonstrating that established government contractors with existing security clearances can rapidly expand access while high-profile partnerships languish. Palantir now holds contracts across NHS, police, military, and financial regulatory data — creating a single vendor concentration risk that privacy groups have consistently opposed but ministers continue to ignore. The FCA contract terms, including data retention and access controls, remain undisclosed.

Anthropic's preliminary injunction hearing on 24 March challenges the Department of Defense's supply chain designation that prohibits federal agencies and contractors from using its AI models, as covered by The Guardian and Lawfare. The designation followed Anthropic's refusal to allow its technology to support autonomous weapons systems, representing the administration's use of procurement restrictions as coercive tools rather than security-based exclusions. The legal question is whether the executive branch can impose broad bans on commercial AI services based on a company's refusal to support specific military applications rather than on security vulnerabilities or foreign ownership.

This differs fundamentally from cases like the TikTok ban, which rested on foreign adversary concerns, as noted in Lawfare's archive discussion. The Pentagon's action suggests a new doctrine where AI companies must either accept military end-uses or face exclusion from all federal business, including civilian research and administrative functions. If upheld, this establishes precedent allowing the executive to punish private sector AI governance decisions through procurement leverage without congressional authorisation.

Analysis from the Institute for AI Policy and Strategy concludes that H200 exports to China would 'substantially boost Chinese frontier AI development and deployment capabilities' for military applications, with verification requirements being 'unenforceable' once chips enter China, according to IAPS research. This assessment directly contradicts BIS claims that end-user controls can prevent diversion to military uses. The research argues that semiconductor manufacturing equipment controls have proven more effective at maintaining U.S. advantages, as detailed in separate IAPS analysis, because equipment is harder to conceal and repurpose than individual chips.

The gap between stated export control policy — preventing Chinese military AI advancement — and enforcement reality suggests either wilful blindness or capacity limitations at BIS. Industry has consistently lobbied for relaxed chip export rules citing competitiveness concerns, and the H200 case indicates those arguments may be prevailing over security assessments. The Commerce Department has not publicly responded to the IAPS analysis or adjusted export license policies.