On March 6, 2026, the General Services Administration (GSA) proposed GSAR 552.239-7001, titled "Basic Safeguarding of Artificial Intelligence Systems," a first-of-its-kind contract clause imposing dedicated AI-specific safeguarding requirements for federal procurement.[1] The proposed clause would impose contractually binding obligations governing the development, deployment, and management of AI systems used in or supplied under federal contracts, and applies broadly to all AI use in contract performance—whether AI is provided directly to the government, embedded in contractor workflows, operated by subcontractors, or licensed through third-party vendors.[1][4]
Who's Involved
The GSA is the primary actor, though the clause affects all contractors relying on GSA MAS (Multiple Award Schedule) contracts who either sell or use AI in contract performance.[1][8] The clause's requirements flow down to subcontractors and "Service Providers" (defined as entities that provide, operate, or license AI systems but are not contract parties), creating obligations for commercial AI vendors such as OpenAI, Google, and Microsoft.[4] The proposed clause follows "the government's very public breakup with Anthropic."[4]
Context and Timeline
The clause reflects growing federal concern about data security, supply chain risk, and the opacity of commercial AI systems.[1] It was issued pursuant to OMB Memo M-25-22, Driving Efficient Acquisition of Artificial Intelligence in Government, which directed agencies to include specific terms in AI contracts.[12] GSA originally requested public comments by March 20, 2026, but extended the deadline to April 3, 2026 in response to industry requests, and announced it would delay implementation to a later MAS Refresh (Refresh 32 rather than Refresh 31).[1][6][8]
Why It's Newsworthy
The clause represents the most comprehensive federal attempt to define contractor obligations for AI deployment in government contracts to date.[4] Key provisions include prohibiting contractors from using government data to train AI models, requiring segregation and deletion of government data, mandating use of only "American AI Systems," and granting the government broad licensing rights to AI systems and outputs.[1][5][8] Critically, the clause includes an order of precedence provision that overrides conflicting commercial terms and conditions, meaning GSA requirements supersede standard vendor EULAs and licensing agreements—potentially forcing contractors to choose between complying with the clause or ceasing use of major commercial AI platforms during contract performance.[4][5]