In late February 2026, OpenAI agreed to let the US military use its AI models in classified settings, and a wave of ChatGPT users reacted by deleting the app. App analytics firm Sensor Tower measured a 295% day over day jump in US uninstalls of the ChatGPT mobile app on February 28. Here is what the deal actually covered, where that number came from, how the companies and the public responded, and which alternatives people turned to.
What led up to the deal
The story started with Anthropic. On February 26, 2026, its chief executive Dario Amodei published a statement saying the Pentagon wanted AI providers to accept any lawful use of their models. Anthropic said it supported national defense work but would not drop two limits from its contracts: no use for mass domestic surveillance and no fully autonomous weapons that pick and engage targets without a human involved. It said it would hold that position even if it were removed from defense systems or labeled a supply chain risk.
The government then moved to cut Anthropic off from federal work and designated it a supply chain risk. Anthropic sued in March 2026, and on March 26 a federal judge issued a preliminary injunction temporarily blocking the designation while the case continues.
What OpenAI agreed to
OpenAI announced its agreement on February 28, saying it had been reached the day before. According to OpenAI, the deal allows its models to run in classified environments under a set of conditions:
- Red lines: no mass domestic surveillance, no directing autonomous weapons systems, and no high stakes automated decisions such as social credit systems.
- Cloud only: no deployment on edge devices and no models with safety guardrails removed.
- Oversight: OpenAI keeps control of its safety systems and involves cleared engineers and researchers.
- Exit clause: OpenAI says it can end the contract if the government breaks its terms.
After criticism, OpenAI updated its post on March 2. It added contract language stating the system shall not be intentionally used for domestic surveillance of US persons, including through commercially bought personal data, and said Pentagon intelligence agencies such as the NSA would not use the service without a new agreement. OpenAI also said it asked for the same terms to be offered to all AI companies.
The uninstall numbers and who measured them
The figures come from Sensor Tower, which estimates app activity from market data rather than from OpenAI’s own records. Its key findings for the US market:
| Metric (US, Sensor Tower estimates) | Change |
|---|---|
| ChatGPT uninstalls, Feb 28 | Up 295% day over day (typical: about 9%) |
| ChatGPT average daily uninstall rate, Feb 28 to Mar 3 | Up 200% versus the prior 30 days |
| ChatGPT downloads, Feb 28 and Mar 1 | Down 13%, then down 5% day over day |
| Claude downloads, Feb 27 and Feb 28 | Up 37%, then up 51% day over day |
| ChatGPT one star App Store reviews, Feb 28 | Up 775% day over day |
Sensor Tower also reported that Claude reached No. 1 in its US App Store category on February 28 and held the spot for four days, with ChatGPT at No. 2. Google Gemini downloads rose 9% day over day across February 27 and 28, compared with a 2% daily average before.
How people and companies reacted
Reaction split along familiar lines. Critics argued that putting commercial AI in classified military systems raised surveillance and weapons risks no matter what the contract said, and some users expressed that by deleting the app and leaving low ratings. Supporters said the US military will use AI either way and that a contract with written limits and an exit clause is better than none. OpenAI chief executive Sam Altman acknowledged the criticism, and the March 2 amendments were presented as a response to it.
On the other side, Anthropic’s refusal drew support from users who agreed with its two red lines, while the administration treated its stance as a reason to cut ties. The legal fight over the supply chain risk label has played out separately in federal court.
Alternatives users considered
Anthropic’s assistant saw the largest download gains in the days after the deal.
Built into Android and Google apps, with a smaller download bump.
Options such as Microsoft Copilot, Perplexity and Mistral’s Le Chat.
Open models that run on your own computer, for maximum privacy.
FAQ
Did ChatGPT really lose 295% of its users?
No. The 295% figure describes the rise in daily US uninstalls compared with the previous day, as estimated by Sensor Tower. It is not a share of total users.
What can the military use OpenAI’s models for?
OpenAI says use is limited to cloud deployments in classified settings and excludes mass domestic surveillance, directing autonomous weapons and high stakes automated decisions.
Why did Anthropic say no?
Anthropic would not accept unrestricted lawful use and insisted on keeping bans on mass domestic surveillance and fully autonomous weapons.
Is the dispute over?
Not entirely. A federal judge temporarily blocked Anthropic’s supply chain risk designation in March 2026 through a preliminary injunction, which is not a final ruling on the case.
The verdict
The uninstall spike was a short, sharp signal of how many users care about how AI companies work with the military. The numbers are third party estimates of daily change, so read them as a measure of sentiment. Whether to stay with ChatGPT or switch depends on how you weigh the contract limits OpenAI published against your own views on military AI.
Updated October 2026: rewritten as a neutral explainer on the OpenAI defense agreement, the Sensor Tower uninstall data, the Anthropic dispute and the alternatives users considered.
Featured photo: Aerps.com on Unsplash.
About this article: GeekBlog covers U.S. technology news, AI, phones, smartwatches and gaming. Every story is written and checked under our Editorial Policy. Spotted a mistake or have a story tip? Contact our editors.

