OpenAI does deploy its models to U.S. government classified networks, but not under a product called 'GPT Gov.' A viral social media post claimed OpenAI's head of national security policy accidentally revealed the existence of a classified government model, but fact-checking reveals the product name doesn't exist in any OpenAI document, the actual government product is called ChatGPT Gov, and the classified-network deployment has been public for seven months.

What Actually Happened With That Viral Post?

On September 26, a post on X quoted Sasha Baker, OpenAI's Head of National Security Policy, describing "a government specific model that we call GPT Gov that we've just introduced into the classified networks." The post framed this as a major slip that "almost nobody caught." However, searching OpenAI's public statements, newsroom announcements, and government records reveals no product called "GPT Gov" anywhere. The string "GPT Gov" returns zero results in OpenAI-authored documents; every government-related product reference points to "ChatGPT Gov" instead.

Sasha Baker

Baker is a real executive with an accurate title. She took the role of Head of National Security Policy in August 2024 after serving as Acting Under Secretary of Defense for Policy. She does speak publicly about government AI work, including appearances at Semafor's Next 3 Billion event on September 22, 2026, where she discussed U.S.-China AI safety talks, and on September 23 when she announced OpenAI's expansion of its Daybreak cyber-defense program to Ukraine. Neither appearance involved a model called "GPT Gov," and the most likely explanation is that the post either misheard or compressed "ChatGPT Gov" into a cleaner-sounding model name.

What Government Products Does OpenAI Actually Sell?

OpenAI's government offerings exist in three distinct tiers, and the confusion between them is what made the viral post sound like a revelation. Understanding the difference matters because they operate at different security levels and serve different purposes.

• ChatGPT Gov (Commercial Cloud): Launched January 28, 2025, as a self-hosted deployment on Microsoft Azure Government cloud for agencies handling non-public sensitive data. It includes enterprise features like single sign-on, admin consoles, user management, and shared custom GPTs, with the agency controlling the data boundary.
• ChatGPT Enterprise for Government (FedRAMP Moderate): Live since August 4, 2026, this tier uses the GPT-5.6 series models (Sol, Terra, and Luna) for government workflows within FedRAMP's Moderate boundary, which covers controlled unclassified information. This is not classified data.
• Classified Network Deployment (DoD Agreement): A separate agreement signed February 28, 2026, between OpenAI and the Department of War to deploy OpenAI models on classified cloud networks for lawful operational use. This is the only tier that actually involves classified networks.

These three tiers are routinely collapsed into one another in casual discussion, which is why a sentence about classified networks sounds like breaking news when it's actually describing a seven-month-old procurement story.

When Did the Classified Network Agreement Actually Become Public?

The classified-network deployment is not a rumor or a slip. Reuters reported it on February 28, 2026, quoting OpenAI CEO Sam Altman directly, describing an agreement to deploy the company's AI models on classified cloud networks. OpenAI published its own account titled "Our agreement with the Department of War" the following day. By May 1, the Department of War had published a broader release titled "Classified Networks AI Agreements" covering eight frontier AI firms, not just OpenAI.

Reuters

The program faced public criticism over what the original agreement permitted. The BBC reported on March 3 that OpenAI had "agreed changes" to a deal critics called "opportunistic and sloppy" regarding use of its technology in classified military operations. Altman told staff that the government owns operational decisions and told a wider audience in early April that he had "miscalibrated" the mood of distrust around the deal. By August, the practical result was clear: GPT-5.6 models available to government users through ChatGPT Enterprise at the FedRAMP Moderate security level.

How to Distinguish Between OpenAI's Government Tiers

• Check the Security Level: ChatGPT Gov handles non-public sensitive data at an agency's own accreditation level. ChatGPT Enterprise for Government handles controlled unclassified information under FedRAMP Moderate. Only the DoD classified network agreement involves actual classified data.
• Look at the Deployment Model: ChatGPT Gov is self-hosted on Azure Government cloud, meaning the agency controls the infrastructure. ChatGPT Enterprise for Government runs on FedRAMP-certified infrastructure. The classified network deployment uses the Department of War's own classified cloud networks.
• Verify the Timeline: ChatGPT Gov launched in January 2025. ChatGPT Enterprise for Government went live in August 2026. The classified network agreement was signed in February 2026 and announced publicly the same day.
• Confirm the Models Used: ChatGPT Gov uses OpenAI's frontier models in a self-hosted setup. ChatGPT Enterprise for Government specifically uses GPT-5.6 Sol, GPT-5.6 Terra, and GPT-5.6 Luna. The classified network agreement deploys OpenAI models on DoD infrastructure without specifying a single model name.

Why Does the Naming Confusion Matter?

The interesting failure here is not that OpenAI does government work. It obviously does, and that work is documented, announced, and subject to public scrutiny. The failure is that a naming conflation traveled further in a single day than the underlying facts did in seven months. A real executive, a plausible topic, and a product name that does not exist in any OpenAI document created the appearance of a major revelation when the actual story was already public.

This pattern highlights how quickly misinformation can spread in AI policy discussions, even when the underlying facts are already on the record. The classified-network deployment is a legitimate procurement story with Reuters reporting, an OpenAI statement, and a Department of War press release behind it. Nothing about it was "let slip this week." The only thing that was new was the incorrect product name.