The General Services Administration (GSA) introduced a landmark acquisition regulation, GSAR 552.239-7001, titled “Basic Safeguarding of Artificial Intelligence Systems.” This proposed clause represents the federal government’s first comprehensive attempt to standardize Artificial Intelligence (AI) procurement through binding contract terms rather than voluntary guidance. The clause introduces stringent requirements regarding data ownership, domestic sourcing, and risk management that diverge significantly from standard commercial practices.
The clause is designed to operationalize recent executive policy, specifically requirements for “American AI” and “Unbiased AI Principles.” It fundamentally shifts the intellectual property (IP) landscape by asserting government ownership over AI outputs and “Custom Developments,” while strictly prohibiting the use of government data for model training. While GSA initially intended to incorporate this clause into the Multiple Award Schedule (MAS) via Solicitation Refresh 31, the agency has deferred inclusion to Refresh 32 following industry feedback, with the public comment period extended through April 3, 2026. The immediate market reaction highlights significant concerns regarding the broad definition of “Service Providers,” which extends liability to upstream vendors who may not be direct signatories to government contracts.
Regulatory Context and Scope of Applicability
The introduction of GSAR 552.239-7001 marks a pivot from broad AI safety guidelines to prescriptive contractual obligations. The clause is slated for inclusion in all GSA Schedule solicitations and contracts for “Artificial Intelligence capabilities.” It aligns GSA procurement with the “Unbiased AI Principles” outlined in the July 2025 Executive Order, Preventing Woke AI in the Federal Government, and Office of Management and Budget (OMB) Memorandums M-25-22 and M-26-04 (Federal News Network, GSA’s new AI clause drives contractors to sound the alarm — March 24, 2026).
A critical aspect of the clause is its expansive scope regarding the supply chain. It applies not only to prime contractors but also to “Service Providers,” defined as any entity that “directly or indirectly provides, operates, or licenses an AI system” used in contract performance (Crowell & Moring, AI for Government: 7 Days for Contractor Comments on…). This definition explicitly captures entities that may not be formal subcontractors, such as commercial cloud providers or third-party model developers. Consequently, prime contractors are contractually responsible for ensuring these upstream providers comply with the clause’s strictures, creating a “flow-down” liability structure that extends deep into the commercial AI ecosystem (Hogan Lovells, UPDATE: March 20, 2026 Deadline for Comments…).
Defining "AI Systems" and Procurement Boundaries
The clause’s applicability hinges on the definition of “AI capabilities,” a term that legal analysts note is currently undefined in the draft, creating ambiguity regarding the full scope of covered products (Holland & Knight, GSA’s Proposed AI Clause: A Deep Dive…). However, the clause incorporates the definition of “AI System” from the Advancing American AI Act (40 U.S.C. § 11301), which broadly includes any software or tool operating in whole or in part using machine learning algorithms (Wiley Rein, GSA Delays Rollout…).
Inclusions and Exclusions:
Captured Categories: The clause captures “all AI systems used in performance of the contract,” encompassing not only deliverables sold to the government but also internal tools used by the contractor, such as embedded AI coding assistants, decision-support tools, and generative models (Gibson Dunn, GSA AI Procurement Rules…).
Exclusions: GSA Directive 2185.1C suggests an exclusion for “common commercial products with embedded AI functionality not primarily used for AI purposes,” though this exclusion is not explicitly detailed in the summaries of the GSAR clause itself, creating potential conflict between internal policy and external contract terms (GSA, CC050878 CIO 2185.1C Use of Artificial Intelligence at GSA).
“American AI” Restriction: The clause imposes a strict domestic sourcing requirement, mandating the use of “American AI Systems,” defined as systems “developed and produced in the United States.” It expressly prohibits AI components “manufactured, developed, or controlled by non-U.S. entities.” This requirement lacks a clear test for what constitutes “produced” in the context of global software development, potentially disqualifying systems trained on global data or utilizing open-source components from international contributors (Holland & Knight, GSA’s Proposed AI Clause…).
Vendor Obligations: Transparency, Documentation, and Data Rights
The clause imposes rigorous transparency and data handling requirements that supersede standard commercial terms.
Transparency and Disclosure: Contractors must disclose all AI systems used in contract performance to the contracting officer within 30 days of award. This includes disclosing whether models have been modified to comply with non-U.S. regulatory frameworks (e.g., the EU AI Act) (Gibson Dunn, GSA AI Procurement Rules…). Furthermore, contractors must provide concurrent access to major model updates for 30 days before discontinuing old versions (GSA, Proposed Government AI System Terms…).
Data Rights and Usage Restrictions: The clause fundamentally alters data ownership:
Government Ownership: The government retains full ownership of all “Government Data” (inputs and outputs) and “Custom Developments,” defined broadly to include model fine-tuning, configurations, and associated workflows.
Training Prohibition: Vendors are explicitly prohibited from using Government Data to train, fine-tune, or improve their AI models for any other customer or commercial purpose.
“Eyes Off” Handling: Contractors must implement “eyes off” procedures, restricting human review of government data to instances that are “strictly necessary,” with all access logged (Baker Botts, GSA’s New AI Clause…).
Difference from Existing Regulations: Unlike standard FAR rights where contractors typically retain ownership of data generated during performance while granting the government a license, this clause assigns ownership of “Custom Developments” (e.g., fine-tuned weights) directly to the government. Crucially, the clause contains an order-of-precedence provision that overrides any conflicting commercial End User License Agreements (EULAs) or Terms of Service (TOS), effectively nullifying vendor-imposed safety restrictions or data usage rights (Lawfare, The GSA’s Draft AI Clause Is Governance by Sledgehammer).
Risk Management and the "Unbiased AI" Mandate
Risk management under GSAR 552.239-7001 is heavily focused on ideological neutrality and operational security.
“Unbiased AI Principles”: Contractors must make “commercial efforts” to ensure AI systems align with “Unbiased AI Principles.” This mandates that systems be “truthful,” prioritize “historical accuracy” and “scientific inquiry,” and operate as “neutral, nonpartisan tools.” The clause explicitly prohibits the manipulation of responses in favor of “ideological dogmas such as Diversity, Equity, Inclusion (DEI)” (The White House, Preventing Woke AI in the Federal Government). Additionally, the system must “acknowledge uncertainty” where information is contradictory.
Operational Constraints:
Anti-Refusal: The AI system must not refuse to produce outputs based on the contractor’s or Service Provider’s “discretionary policies” (e.g., commercial trust and safety filters that exceed legal requirements).
Incident Reporting: Contractors must report confirmed or suspected incidents to CISA and the contracting officer within 72 hours, provide daily updates, and preserve forensic artifacts for 90 days (Crowell & Moring, AI for Government…).
Verification, Audits, and Compliance Enforcement
The clause grants the government extensive oversight powers that function as continuous audit mechanisms.
Audit and Assessment: The government reserves the right to conduct “automated assessments” of deployed AI systems at any time using its own benchmarks to test for bias, truthfulness, safety, and “unsolicited ideological content.” These assessments may use “undisclosed methodologies,” meaning vendors may not know the criteria against which their systems are being tested (Lawfare, The GSA’s Draft AI Clause…).
Documentation Obligations: Upon request, contractors must provide confidential documentation demonstrating compliance, including:
- System documentation consistent with the NIST AI Risk Management Framework.
- System decision-making logic and operational parameters.
- “System cards” or equivalent transparency artifacts detailing known biases and limitations.
- methodologies used to test for noncompliance with “Unbiased AI Principles” (GSA, Proposed Government AI System Terms…).
Enforcement and Liability: If an AI system is found non-compliant, the government may suspend its use until issues are addressed. If a contract is terminated for failure to comply with Unbiased AI Principles, the contractor may be liable for “reasonable decommissioning costs.” Legal analysts warn that because compliance is “material to contract eligibility,” non-compliance could trigger False Claims Act (FCA) liability, particularly given the difficulty of certifying the compliance of third-party “Service Providers” (Baker Botts, GSA’s New AI Clause…).
Conclusion: Strategic Implications for the Federal Supply Chain
GSAR 552.239-7001 signals a decisive shift in the federal AI marketplace from a “commercial-first” approach to a “government-unique” compliance regime. By overriding commercial safety policies, asserting ownership over fine-tuned models, and enforcing undefined “American AI” sourcing requirements, the clause creates significant barriers to entry for global technology firms.
The tension between the government’s demand for “Unbiased AI” (defined as excluding DEI-related guardrails) and commercial safety standards (which often include such guardrails) forces vendors into a bifurcated market strategy. Contractors must immediately assess their exposure by inventorying all AI tools used in performance and evaluating their upstream Service Providers for compliance. With the clause expected to be incorporated into MAS contracts via Refresh 32, vendors have a limited window to prepare for these “take-it-or-leave-it” terms or risk losing eligibility for federal awards.