Research published criticisms, identified implementation failures, and counterarguments to the U.S.

The U.S. federal AI strategy—spanning executive orders, the National AI Initiative, and the 2025 Trump administration's America's AI Action Plan—has faced repeated implementation shortfalls, policy reversals, funding cuts, and structural critiques that undermine its effectiveness.[1][2]

Documented evidence from GAO reports, academic assessments, and contemporaneous analyses reveals persistent bureaucratic capacity problems, politicized disruptions to research funding, and contradictory policy signals. These create credible risks of stalled progress, uneven outcomes, and eroded public/international trust. Below is a structured overview of the most credible risks and disconfirming evidence, organized by key failure modes.

Bureaucratic Implementation Shortfalls: Mandates Routinely Go Unmet Due to Weak State Capacity

A systematic 2022–2023 empirical review of three core pillars of U.S. AI governance (AI Leadership EO 13859, Trustworthy AI EO 13960, and the AI in Government Act) found that fewer than 40% of 45 binding legal requirements could be verified as implemented, with roughly 27% fully implemented overall.[2][2]

• Nearly half or more of agencies failed to publish required AI use-case inventories despite demonstrable machine-learning deployments; when published, data was often incomplete, inaccurate, or duplicative (GAO confirmed similar issues in 2023 inventories across 23 CFO Act agencies).[3]
• ~88% of agencies subject to Agency AI Plan requirements under the Leadership EO failed to submit them by late 2022.
• GAO’s 2023–2025 reviews showed only partial compliance: 10 of 23 agencies fully implemented applicable requirements, while others (plus OMB and OPM) fell short on acquisition guidance, workforce planning, and inventory standards. Generative AI usage surged ninefold (2023–2024), but agencies cited ongoing compliance, budget, and policy-update barriers.[4]

This pattern indicates that even well-intentioned mandates falter without dedicated technical staff, clear delineation of covered AI, and enforcement mechanisms—implying that new entrants or agencies attempting rapid scaling will encounter the same execution gaps unless they invest heavily in internal capacity first.

Funding Disruptions and Research Undermining: Core Research Infrastructure Targeted Amid Strategy Rollout

The 2025 America’s AI Action Plan assigns NSF lead roles in AI research labs, testbeds, workforce development, and the National AI Research Resource—yet the same administration canceled >1,500–1,600 NSF grants worth >$1 billion in 2025, many supporting foundational AI, STEM education, or related work, often via political/DEI criteria or form letters.[1][5]

• Cancellations included projects on AI bias mitigation, deepfake detection, AI literacy, and broader computational research; staff in relevant divisions were terminated and peer review processes politicized.
• Billions in congressionally authorized funds were reportedly impounded, and a federal court issued a preliminary injunction reinstating some UC-system awards.
• Industry and analysts noted these cuts directly contradict calls for increased public-private AI R&D partnerships.

Skeptics argue this self-undermines competitiveness claims: defunding the primary public funder of long-term, interdisciplinary AI research while accelerating private-sector deployment creates a dependency on industry without independent safeguards or talent pipelines—raising risks that “innovation” rhetoric masks hollowed-out foundational capacity.

Policy Volatility and Reversals: Strategic Direction Shifts Abruptly with Administrations

President Trump rescinded Biden’s EO 14110 (“Safe, Secure, and Trustworthy AI”) shortly after taking office in January 2025, replacing it with EO 14179 (“Removing Barriers to American Leadership in AI”) and the July 2025 America’s AI Action Plan emphasizing deregulation, “objectivity,” and rapid adoption.[6]

• Prior EOs and guidance (e.g., OMB M-24-10) were superseded by more flexible, agency-driven memos (M-25-21), reducing centralized oversight but increasing implementation variability.
• Historical pattern: Requirements from one administration are often not continued, leading to “non-responses” and wasted agency effort.

This volatility signals to skeptics a fundamental flaw—AI governance treated as partisan rather than enduring infrastructure—making long-term planning or cross-administration commitments unreliable and deterring sustained investment.

Governance Fragmentation and Capacity Gaps: Uneven Agency Readiness Produces Patchy Protections

Recent guidance shifts greater responsibility to individual agencies for AI governance and compliance plans, but agencies vary widely in technical resources, staffing, and expertise.[7]

• Smaller or resource-constrained agencies (often those serving vulnerable populations) struggle most; well-funded ones can comply while others cannot.
• OECD analysis of government AI use cases highlights systemic barriers: skills gaps, data-access difficulties, and many initiatives remaining stuck in pilots rather than scaling.
• Legacy systems, siloed data, and diffused ownership block scaling of promising pilots (noted in 2026 analyses of permitting and other workflows).

For competitors or new federal programs, this implies that “agency discretion” models favor incumbents with existing capacity and risk creating a two-tier system where protections and benefits are inconsistently applied.

Regulatory Patchwork, Preemption Risks, and Sector-Specific Concerns: External and Internal Contradictions Compound Uncertainty

Absence of comprehensive federal legislation has driven state-level activity (>800 bills since 2019), creating compliance complexity that the federal strategy seeks to address via preemption threats or funding conditions—yet these face legal uncertainty and potential court challenges.[8]

• Brookings and other critics highlight contradictions in the 2025 Action Plan: heavy-handed federal interference in AI “objectivity”/speech moderation while pursuing broad deregulation; failure to address baked-in data biases (e.g., in credit/finance); and a “try-first” culture in healthcare despite evidence that most LLM studies lack real-world validation (only ~5% in one review incorporated patient-care data) and rigorous prospective evaluation.[1]
• Dismantling oversight bodies (e.g., CFPB) raises questions about monitoring AI in high-stakes areas like credit decisions.

Skeptics contend these reveal deeper flaws: an approach that prioritizes speed and private leadership without commensurate investment in evaluation science, bias mitigation, or coordinated governance, potentially leading to harms, legal exposure, or international credibility loss.

Overall, the most credible risks center on execution gaps rooted in bureaucratic capacity, research-funding instability, and policy inconsistency rather than the absence of strategy documents. These disconfirming factors suggest that federal AI initiatives have repeatedly under-delivered on transparency, scaling, and sustained momentum, with reversals and cuts actively eroding supporting institutions. Entities seeking to engage with or compete in this space should prioritize independent technical capacity, diversified funding, and contingency planning for abrupt shifts. Additional primary-source verification (e.g., latest GAO follow-ups or court outcomes on grant reinstatements) would further strengthen assessments of ongoing developments.

---

Recent Findings Supplement (June 2026)

Federal-state tensions escalated sharply in late 2025–early 2026 through executive actions aimed at preempting or challenging state AI laws.[1]

The Trump administration’s December 2025 executive order (and related directives) established a DOJ AI Litigation Task Force (announced January 10, 2026) to challenge state laws conflicting with federal policy, directed Commerce to evaluate “onerous” state rules for potential federal funding cuts (e.g., BEAD broadband funds by March 2026), and signaled broader preemption efforts, including possible FCC proceedings. This created immediate legal uncertainty, with states and critics arguing Tenth Amendment violations, overreach without a replacement federal standard, and risks to innovation from patchwork compliance burdens or funding threats.[2]

• Multiple states continued advancing or defending their AI laws despite federal pressure; existing state privacy/AI statutes remain in effect where not directly conflicted.
• Critics (including state legislators) highlighted bypassed congressional processes and lack of national standards as destabilizing.
• No comprehensive federal AI privacy or governance statute has emerged to replace state actions.

Implications for competitors/entrants: Expect prolonged litigation and compliance fragmentation; companies must monitor state-federal clashes closely, as preemption attempts may face court challenges or congressional pushback, favoring those with flexible, multi-jurisdictional strategies or strong state-level relationships.

A March 2026 GAO report identified significant gaps in OMB’s government-wide AI guidance on privacy risks.[3]

Expert panels convened by GAO identified 13 privacy-related challenges with AI use (e.g., revealing sensitive information from datasets, performance tradeoffs when removing data for privacy, lack of enabling technologies). OMB guidance (as of January 2026) addressed only a subset—such as workforce skills and scaling AI with protections—while failing to specify known privacy risks agencies should consider or fully tackling eight of the ten selected challenges. Recommendations include updated guidance and interagency sharing via Chief AI Officer or Privacy Councils.[4]

• Risks include unintended disclosure of PII or sensitive data in federal AI applications.
• OMB provided no comments on the draft report.

Implications: Agencies and contractors face heightened compliance and audit risks; entrants should prioritize privacy-enhancing technologies (PETs) and robust documentation to differentiate amid guidance shortfalls.

NIST’s March 9, 2026 report (NIST AI 800-4) details persistent challenges in monitoring deployed AI systems.[5]

As AI integrates into government and commercial applications, real-world monitoring faces gaps, barriers, and open questions across categories like performance drift, security, and accountability. The report organizes expert-identified issues to guide future research rather than providing immediate solutions.

• Emphasizes demand for ongoing oversight beyond initial deployment.
• No specific federal implementation failures cited, but highlights systemic evaluation difficulties.

Implications: Vendors offering monitoring, auditing, or MLOps tools have opportunities; government contractors must build in post-deployment capabilities to meet emerging expectations.

The Anthropic–DOD/Pentagon dispute (escalating February–May 2026) represents a high-profile implementation clash between company ethics policies and federal demands.[6]

Anthropic refused unrestricted use of its Claude models for autonomous weapons or mass domestic surveillance (consistent with prior contract terms), leading to contract termination (estimated $200 million value), a presidential directive to cease all federal use, and DOD designation of the company as a “supply chain risk” (typically reserved for adversaries). Anthropic sued in March 2026; a district court issued a preliminary injunction (later impacted by appeals court rulings), framing actions partly as potential First Amendment retaliation. Other firms (e.g., OpenAI) filled gaps with new deals.[7]

• Highlights ethical/national security tensions in defense AI procurement.
• Resulted in blacklisting effects on contractors and public scrutiny.

Implications: Companies with restrictive use policies risk federal contract losses or blacklisting; those aligned with unrestricted defense applications gain advantages. This underscores risks of value misalignment in government AI partnerships.

Brookings analyses (April–May 2026) show federal AI spending surging but heavily skewed toward Defense, with broader adoption hurdles.[8]

Contract data indicate explosive growth (obligated funds to $7.2 billion; potential awards to $91.8 billion by early 2026), dominated by DOD (nearly 99% of potential value). Other agencies remain marginal. Adoption faces talent shortages, trust deficits, and possible disruption from initiatives like DOGE. Privacy gaps and monitoring challenges compound these.[9]

• Market shows maturation (larger vendors, firmer pricing) but remains experimental in non-defense areas.
• Ethical clashes (e.g., Anthropic case) illustrate contested use cases.

Implications: Non-defense entrants face limited opportunities unless targeting infrastructure or specific civilian pilots; success requires navigating talent gaps and proving trustworthiness amid uneven federal priorities.

Overall, recent evidence points more to structural frictions—guidance shortfalls, intergovernmental conflicts, and ethical-procurement clashes—than outright program cancellations or funding reversals. Skeptics emphasize inadequate privacy oversight, preemption without standards creating uncertainty, and defense-centric approaches that sideline balanced governance or civilian innovation. Additional primary sources (e.g., court filings or agency responses) would further clarify litigation outcomes.