OpenAI and the U.S. General Services Administration announced a government agreement on September 10, 2026 that changes how federal, state, local and tribal agencies can access OpenAI services. The offer combines a zero-dollar license fee for eligible access, 50% off usage and additional support for public-sector cyber defenders.
What the OneGov offer actually changes
OpenAI says the agreement provides free access to eligible government users where the normal license fee is $15 per user per month, while usage is discounted by 50%. The agreement is described as multi-year, running from October 1, 2026 through December 31, 2028.
The key point is that the offer is not simply “free ChatGPT for government.” Usage still has a cost, and OpenAI describes separate support, usage estimates, spend controls, onboarding and FinOps guidance. Eligibility and purchasing details also need to be confirmed with the government team.
Why the 27-month term matters
A fixed period through the end of 2028 gives agencies a clearer planning window than a short pilot. Procurement, training and security review can take months, so a longer commercial framework can make it easier to build an adoption plan instead of treating AI access as a temporary experiment.
It does not guarantee that every agency will deploy the same way. Agency-specific security requirements, approved services and internal policies still control how the tools can be used.
What OpenAI says about data protection
OpenAI states that ChatGPT Enterprise does not use business data, including inputs or outputs, to train or improve OpenAI models. The announcement says participating organizations retain the protections associated with the eligible services they choose. Agencies should still review the exact service, contract terms and data controls before putting sensitive government information into a workflow.
The cyber-defender part is separate and important
OpenAI says the agreement includes expanded support and access for public-sector cyber defenders. That fits with its September 3 Daybreak initiative, where OpenAI announced a $1 billion commitment in subsidized access, training, technical support and partnerships for frontline defenders.
OpenAI describes Daybreak use cases such as reviewing legacy code, analyzing suspicious activity, identifying and validating vulnerabilities, prioritizing risks, and developing and testing fixes. These are security workflows, but they still require human review and organizational controls before changes reach production systems.
What agencies should measure
The useful measures are not just the number of employees with access. Agencies should track actual adoption, usage cost, high-value workflows, review time, security incidents, approved data types and the amount of work that moves from experiment to repeatable process.
The agreement's inclusion of usage estimates and spend controls is especially relevant because a zero license fee can otherwise hide variable usage costs. The economics should therefore be measured as total workflow cost rather than seat price alone.
What is still not public
The announcement does not give a complete list of every eligible agency, all service entitlements, every model limit or the exact usage pricing for every workload. It also does not establish that all government workloads are appropriate for the same ChatGPT configuration. Those details should be checked during procurement.
The practical takeaway
OneGov lowers an important adoption barrier, but the real change is the combination of access, usage discounts and operational support. Agencies that treat the program as a governed deployment—with clear data rules, spend controls, security review and measurable workflows—will get more useful information from the offer than agencies that measure success only by account creation.