Announced 28 Sept 2026 · Sources checked
What did GSA put in RGO-2026-01?
RGO-2026-01 is the Senior Procurement Executive’s class-deviation memorandum for the Revolutionary GSAR Overhaul. The bundled PDF on Acquisition.gov includes Part 539 prescription text and the full 552.239-7001 clause labeled “(GSAR Deviation)(Sept 2026).”
539.71 directs contracting officers to insert 552.239-7001 when procuring an LLM, generative assistant, chatbot, agentic system, LLM-enabled document or productivity tool, or similar system where LLM functionality is a material feature and Government Data will be submitted directly to or produced by the LLM.
This is procurement language, not a model release. For a separate federal AI compute story from the same week, see the Genesis Mission industry pledges note.
RGO-2026-01 is a procurement instruction, not a model evaluation. Its force comes from where Attachment A lands in GSAR Part 552 and from which solicitations incorporate the clause before the self-delete date. Vendors should track the clause number and the solicitation’s Section I/K references rather than relying on informal email summaries of “the new AI clause.”
What does Attachment A say about timing?
Attachment A explains that the table tracks each GSAR part’s deviation effective date. Contracting activities may use the new RGO language immediately but must begin using it no later than the listed date.
On the Attachment A pages we inspected as rendered PDF pages, Part 552 shows Effective Date 10/19/2026 and Date Signed 07/16/2026. Part 539 shows an earlier 03/16/2026 row; the LLM clause text itself lives under Part 552.
FedScoop and other trade outlets described the AI clause’s mandatory horizon as 19 October after the memo circulated; the Attachment A Part 552 row is the primary schedule entry we opened for that date.

When does the clause self-delete?
Unless the contracting officer says otherwise, the clause imposes no obligation when LLM use stays inside the contractor’s internal business, back-office, operational, or performance-support systems that are not delivered to or accessed by the Government—even if those systems incidentally process Government Data.
It also self-deletes when LLM functionality is incidental or ancillary to commercial products whose primary purpose is not AI or LLM functionality and the LLM feature is not required in the solicitation or contract.
Flowdown follows roles that design, develop, deploy, or operate the model for Government Data—not authorship alone. That “follow the data” framing matches industry comments summarized in FedScoop’s 1 October explainer.
What does the clause require when it applies?
Prohibited uses of Government Data include training or fine-tuning LLMs, informing advertising or marketing, selling or licensing the data, and transferring it to unauthorized parties. Allowed handling is limited to performing the contract.
Contractors must implement data-handling procedures with access limits, encrypted transmission/processing, audit logging, and related safeguards. Non-persistent architectures that do not retain Government Data are described as satisfying the minimum measures.
The government reserves rights to assess the LLM for bias, truthfulness, safety, and related factors, and to suspend use. Material changes that increase bias or weaken guardrails trigger notice duties in the clause text. Exact day counts and liability caps should be read from the PDF for the order in front of you.
When the clause is active, read the exact duties around training-data use, prompt/log retention, and customer content segregation as written—not as a paraphrase of consumer chatbot privacy blogs. Flow-down to subcontractors and to hosted-model providers is often where teams fail audits. If your architecture sends federal prompts to a third-party frontier API, map each hop to the clause’s verbs before you sign.

What should vendors do before 19 October?
Map whether each GSA bid is procuring LLM functionality as a material feature with Government Data in the loop. If not, document why the self-deletion conditions apply.
Align subcontract flowdown with who actually touches Government Data. Update commercial terms so they do not conflict with the clause’s order-of-precedence language.
Separate product safety policies from contract clauses; a vendor usage policy update does not replace GSAR text on a GSA order.
Calendar the 19 October self-delete: after that date, confirm whether your live solicitations still incorporate the clause by reference or whether a successor instrument replaced it. Do not assume silence means the duties vanished from awarded contracts that already included Attachment A.
Ask counsel whether your standard DPAs and subprocessors already meet the clause or whether you need a federal-specific rider. Product teams should prepare evidence packs: data-flow diagrams, retention timers, and training-opt-out controls that match the text GSA printed.
What did we not do?
We are not giving legal advice. We did not survey contracting officers, compare every prior draft line-by-line, or determine how any specific agency will exercise early adoption. This article reports the RGO-2026-01 PDF and contemporaneous FedScoop explainer we opened.
Common questions
Does this ban contractors from using ChatGPT internally?
Not by default. The clause self-deletes for internal back-office LLM use that is not delivered to or accessed by the Government, unless the contracting officer says otherwise.
Is 19 October the Part 539 date or the Part 552 date?
Attachment A lists Part 552 as 10/19/2026. The clause text is 552.239-7001. Part 539’s own Attachment A row is earlier; read both rows if your work spans prescription and clause text.
Is this final FAR rulemaking?
No. RGO-2026-01 is a class deviation path. The memo says formal rulemaking with notice and comment will follow.
What to remember
GSA’s LLM data-safeguarding language is a dated Part 552 clause with a 19 October self-delete—treat incorporated contracts as the source of duty, and verify flow-down before that calendar mark.
Sources & further reading
How this story was made
Written by Kristian Kostov with AI assistance and checked against the linked sources. Company performance claims are attributed to the company. Analysis reflects AiLookout’s interpretation; we have not independently tested the products discussed. Cover photography is illustrative and does not depict the specific announcement or product.
Our editorial standards





