As part of its ongoing acquisition overhaul under Executive Order 14275, GSA has issued a refreshed version of clause 552.239-7001, Basic Safeguarding of Data within Large Language Model Artificial Intelligence Systems (Sept. 2026). The clause applies whenever GSA procures an LLM, chatbot, agentic system, or LLM-enabled productivity tool where AI functionality is material and government data flows to or from the model — but not to contractors’ internal back-office AI use.
Key provisions include: the government retains full ownership of its data and any custom AI developments, with contractors barred from using that data to train or improve models, for marketing, or for other customers. Contractors must delete government data at contract close, segregate it from other clients’ data, and report security incidents within 72 hours. New transparency requirements require audit trails for agentic/reasoning steps, and GSA reserves the right to benchmark-test deployed models for bias and accuracy, with authority to suspend non-compliant systems. Foreign-control restrictions also bar models subject to adversarial government influence.