The #QuitGPT Movement: What Happened
OpenAI signed a classified Pentagon deal on February 28, 2026. Hours earlier, Anthropic CEO Dario Amodei refused the same contract—walking away from a $200 million agreement on principle.
The symmetry was devastating: one AI CEO said no on ethics; the other said yes for the money.
Key Numbers:
- 295% surge in ChatGPT uninstalls overnight
- 2.5 million users pledged to quit
- Claude hit #1 on App Store for the first time
- $5 billion revenue jump for Anthropic in one month
- 13% drop in US ChatGPT downloads
The Contract: What OpenAI Agreed To
The Pentagon deal gave the Department of Defense access to OpenAI's models on classified networks for "any lawful purpose." OpenAI claims three red lines were carved out:
| Prohibition | Contract Language |
|---|---|
| No domestic surveillance | "The AI System shall not be intentionally used for domestic surveillance of U.S. persons and nationals" |
| No autonomous weapons | "Prohibitions on domestic mass surveillance and human responsibility for the use of force" |
| No high-stakes automated decisions | "The AI System shall not be used for unconstrained monitoring of U.S. persons' private information" |
But the EFF called these "weasel words." The key ambiguities:
- "Intentionally" — The government argues mass surveillance happens "incidentally," not intentionally
- "Consistent with applicable laws" — Laws can change. Executive orders can be rewritten
- "Deliberate" — Intelligence agencies routinely use commercially purchased data to sidestep privacy protections
Katrina Mulligan (OpenAI's head of national security partnerships) confirmed that Defense Intelligence Components (NSA, NGA, DIA) would need "a separate contract modification" to use services.
Anthropic's Refusal: The $200M Walkaway
Anthropic had an existing $200 million Pentagon contract from 2025. When the Pentagon demanded unrestricted access in January 2026, Dario Amodei drew two hard lines:
1. Mass domestic surveillance
"AI-driven mass surveillance presents serious, novel risks to our fundamental liberties"
2. Fully autonomous weapons
"Frontier AI systems are simply not reliable enough to power fully autonomous weapons"
Amodei's exact words: "We cannot in good conscience accede to [the Pentagon's] request."
The Pentagon's budget for autonomous weapons: $13.4 billion in fiscal year 2026 alone.
Defense Secretary Pete Hegseth retaliated immediately:
- Designated Anthropic a "supply-chain risk to national security" (normally reserved for foreign adversaries)
- Directed all federal agencies to stop using Anthropic tools within 6 months
- Accused Amodei of having a **"God-complex""
Altman's Admission: "It Looked Opportunistic and Sloppy"
Three days after signing, Sam Altman published a statement admitting the rollout was rushed:
"We were genuinely trying to de-escalate things and avoid a much worse outcome, but I think it just looked opportunistic and sloppy." "We shouldn't have rushed the announcement."
He renegotiated terms by March 2-3, adding:
- Explicit prohibition on "intentionally" domestic surveillance
- Coverage for "commercially purchased data" (cell phone locations, fitness app info)
- Barred NSA, NGA, DIA without separate contract modification
At an all-hands meeting, Altman told employees the Pentagon "doesn't get to make operational decisions" and that "if the model refuses to do a task, then the government would not force OpenAI to make it do that task."
The Kalinowski Defection
On March 7, 2026, Caitlin Kalinowski—OpenAI's robotics hardware lead—resigned.
Her statement:
"My issue is that the announcement was rushed without the guardrails defined." "Surveillance of Americans without judicial oversight and lethal autonomy without human authorization are lines that deserved more deliberation than they got." "This was about principle."
The resignation reframed the story from consumer boycott to employee accountability. Kalinowski had joined OpenAI in November 2024 from Meta, leading the hardware and robotics group.
The Metrics: User Exodus in Real-Time
| Metric | Before | After Deal |
|---|---|---|
| ChatGPT uninstall rate | 9% (typical) | 295% surge |
| US ChatGPT downloads | Normal | 13% drop |
| 1-star reviews | Baseline | 775% surge |
| Claude US downloads | Normal | 88% surge |
| Claude App Store rank | Top 10 | #1 (first time) |
An Anthropic spokesperson: "Every single day last week was an all-time record for Claude sign-ups."
quitgpt.org launched to organize the boycott. The movement's messaging targeted OpenAI's political ties:
"The President of OpenAI gave Trump $25 million in 2025" "ChatGPT is Trump's biggest 2025 donor"
Greg Brockman (OpenAI President) had donated $25 million to MAGA Inc. super PAC in September 2025.
The Employee Petition
Over 900 former and current OpenAI and Google staffers signed a petition expressing solidarity with Anthropic:
- 266 Google employees
- 65 OpenAI employees
- Hundreds of former employees
The petition opposed AI use for autonomous weapons and mass surveillance, titled "We Will Not Be Divided."
Complete Timeline
| Date | Event |
|---|---|
| June 2025 | Anthropic signed initial Pentagon contract |
| Sept 2025 | Brockman donated $25M to MAGA Inc. super PAC |
| Jan 2026 | Pentagon demanded unrestricted Anthropic access |
| Feb 26, 2026 | Dario Amodei publishes refusal statement |
| Feb 27, 2026 | Anthropic designated "supply-chain risk"; Trump orders federal agencies to stop using Claude |
| Feb 28, 2026 | OpenAI announces Pentagon deal |
| Feb 28, 2026 | ChatGPT uninstalls surge 295%; Claude hits #1 App Store |
| Mar 2, 2026 | Altman admits deal was "sloppy"; renegotiation announced |
| Mar 4, 2026 | ~50 protesters outside OpenAI HQ in SF |
| Mar 5, 2026 | Pentagon officially notifies Anthropic of designation |
| Mar 7, 2026 | Caitlin Kalinowski resigns |
| Apr 10, 2026 | Molotov cocktail attack on Altman's home |
| Apr 12-13, 2026 | Second attack; suspect charged with attempted murder |
The EFF's Core Argument
The Electronic Frontier Foundation published a detailed analysis titled "Weasel Words."
Key critique:
"Secret agreements and technical assurances have never been enough to rein in surveillance agencies. They are no substitute for strong, enforceable legal limits and transparency." "The government is likely attempting to publicly commit to limits in principle, but retain broad flexibility in practice."
The EFF noted the government's existing practices:
- Buying bulk data from brokers (locations, web browsing, associations)
- CBP tapped the advertising ecosystem for tracking
- ICE using device mapping tools
- ODNI proposed centralized data broker marketplace
Community Sentiment
Hacker News reactions:
"I get the feeling they didn't like Anthropic from the start for whatever reason and just found excuses to dump them"
"Never discount the possibility of Hegseth being petty and doing the OpenAI deal with the same terms to imply that Anthropic is being unreasonable"
"It's enheartening to see that leaders at Anthropic are willing to risk losing their seat at the table to be guided by values"
Reddit reactions:
"Hypocritical since OpenAI announced they have zero morals"
"He went full Trump. It underlines that Amodei took the right decision. You definitely don't want to work with people like that"
What It Means
The QuitGPT movement exposed a fault line in AI deployment ethics that technical safeguards alone cannot bridge. OpenAI's contract language—crafted with lawyers and negotiators—still contains ambiguities that privacy advocates consider fatal loopholes.
Amodei's refusal cost Anthropic hundreds of millions in expected revenue and triggered government retaliation. But the user metrics tell a different story: Claude's signups hit all-time records, and Anthropic's revenue reportedly surged.
The uncomfortable question for OpenAI: if contract safeguards are real protections, why did 2.5 million users walk away? And if they're PR cover, what happens when the Pentagon interprets "intentionally" and "consistent with applicable laws" the way the EFF predicts?
Sources: Politico, TechCrunch, Reuters, EFF, Anthropic statement, Fortune, Guardian, NPR, Sensor Tower, Appfigures, Hacker News, Reddit