OpenAI announced a new agreement with the U.S. General Services Administration on September 10, 2026, expanding AI access to federal, state, local, and tribal governments. Eligible organizations can receive ChatGPT Enterprise licenses at no license fee, down from the stated $15 per user per month, and 50% off usage. This is a procurement and deployment move rather than a new model launch, but it could materially change how public-sector organizations begin scaling AI.
The agreement runs for 27 months, from October 1, 2026 through December 31, 2028. OpenAI is also extending support for Daybreak for public-sector cyber defenders to every verified government entity and says Daybreak Blue access, training, and enablement will be offered at a 50% discount. The announcement does not make those capabilities universally unrestricted; eligibility and purchasing still need to be confirmed with OpenAI’s government team.
OpenAI cites early adoption examples including the CDC using ChatGPT to reduce public-health literature reviews that previously took days or months, with most initial reports produced in under 30 minutes and 92% of participating experts reporting productivity gains. Georgia’s Department of Revenue reportedly reduced tax-form digitization from as long as two weeks to 15 minutes. These are customer or partner results reported by OpenAI, not independently reproduced cross-government benchmarks.
Two other examples concern research and public assets. The North Carolina Department of State Treasurer used ChatGPT to identify potentially unclaimed property that could be returned to residents, while Lawrence Livermore National Laboratory used ChatGPT to accelerate a preliminary fusion-research model. These tasks are better understood as AI-assisted retrieval, organization, and candidate generation followed by accountable human review, not as a transfer of administrative decisions to a model.
The package includes buyer guidance, usage estimates, spend controls, onboarding, FinOps guidance, webinars, and an expanded Government Academy. Those details address the practical bottleneck in public-sector adoption. The hard part is rarely opening a chat interface; it is coordinating budgets, data classifications, procurement terms, account management, training, and responsibility across departments. Embedding adoption support in the agreement treats government AI as an operating program rather than a short pilot.
On security, OpenAI says ChatGPT Enterprise does not use business data, including inputs and outputs, to train or improve its models, and that participating organizations retain the protections of the eligible services they choose. That is a provider commitment, not a substitute for agency governance. Each department still needs to define allowed data, connectors, required approvals, and audit retention before it moves real work into the system.
OneGov also places cyber-defense access inside the broader adoption frame. Government IT teams may use the offering to accelerate vulnerability research, malware analysis, and security-tool development, but sensitive environments require separate design for model access, tool permissions, data isolation, human review, and incident response. This article does not turn the announcement’s defensive access into operational attack guidance.
The larger signal is that public-sector AI adoption is being packaged as a combination of price, capability, training, governance, and durable access. Free licenses can lower the trial barrier, but they do not answer questions about data sovereignty, traceability, or public accountability. Agencies should measure task quality, human approval, cost, accessibility, and risk—not just the number of accounts enabled.
