AI Intel: ChatGPT Uninstalls Spike 295%, Anthropic Sues the Pentagon, and the Coding Agent War Heats Up

The AI industry's political crisis just got a lot worse. ChatGPT uninstalls jumped 295% in late February after OpenAI signed its Pentagon contract, and the bleeding hasn't stopped. Meanwhile, Anthropic — the company that walked away from that same deal — is now being blacklisted by the Trump administration and fighting back in court. The fallout is reshaping which AI tools developers trust, and where their API dollars go.

The QuitGPT Exodus Hits 1.5 Million

What started as angry Reddit posts has turned into the largest consumer revolt in AI history. According to Medium reporting and app analytics, roughly 1.5 million users have uninstalled ChatGPT since OpenAI announced its Department of Defense contract in late February. Uninstall rates spiked 295% in the days following the announcement.

The #QuitGPT movement now has organized weekly protests in San Francisco. Sam Altman sent an internal memo on March 2 claiming the contract includes "clearer safeguards preventing the Pentagon from using its models for mass domestic surveillance." That memo leaked to X within hours, and the response was... not reassuring. Multiple OpenAI staffers have quit over the deal.

Here's what matters for developers: user trust is now a competitive moat. If you're building consumer-facing products on OpenAI's API, your users might care which model powers the backend. That wasn't true six months ago. It is now. Claude topped the App Store this week partly because people are voting with their wallets — and their uninstall buttons.

Anthropic vs. The Pentagon: It's Now a Lawsuit

The story behind Anthropic's refusal to work with the Pentagon keeps getting wilder. After walking away from the DoD deal — reportedly over concerns about how the models would be used — Anthropic found itself on the wrong side of the Trump administration. The Guardian reported that defense contractors, plus the State and Treasury departments, have started stepping away from Claude.

Anthropic's response? Sue the Pentagon. The details are still emerging, but the company appears to be challenging the legality of what amounts to a government blacklist for refusing a contract.

Short-term, this is a PR win for Anthropic. The "principled AI company" narrative writes itself, and Claude's user numbers reflect it. Long-term? Less clear. Government contracts are serious money, and being frozen out of federal procurement could hurt Anthropic's enterprise pipeline. For now, though, the developer community is firmly on Anthropic's side.

The practical impact: if you're building for government or defense-adjacent clients, Claude might become a harder sell. For everyone else, Anthropic just became the "safe" choice for teams that care about ethics positioning.

Claude Code Is Winning the Coding Agent War

A DEV Community survey of 500+ Reddit developers landed this week with a verdict that surprised nobody who's actually used both tools: "Claude Code is higher quality but unusable. Codex is slightly lower quality but actually usable."

That was the consensus a month ago. It's shifting. WIRED reported that Codex's user base grew to about 40% of Claude Code's by January 2026, but Claude Code keeps pulling ahead on output quality. The real differentiator isn't raw capability anymore — it's workflow integration. Claude Code treats every task like a collaboration, asking clarifying questions and working incrementally. Codex runs autonomously and hands you the result.

Claude Code Review launched March 9 in research preview, taking a "depth-over-speed" approach compared to Codex's fast inline reviews. For teams doing serious code review, this matters. For quick fixes, Codex still wins on speed.

The API angle here is interesting. Both tools burn through tokens fast — a single Claude Code session can easily consume $5-15 in API calls for a complex refactoring task. If you're running either tool through your own API key, the cost difference between providers adds up quickly. A proxy service like KissAPI can cut those costs significantly, especially for teams running multiple coding agents in parallel.

Claude's Growing Pains: The March 11 Outage

Yesterday's Claude outage was a reminder that rapid growth has consequences. Starting around 2:44 PM UTC on March 11, Claude.ai and Claude Code went down for thousands of users. Login failures, slow responses, and complete unavailability hit at the worst possible time — right in the middle of the US workday.

Anthropic's status page tracked the incident in real-time, which is more transparency than most providers offer during outages. Both issues were resolved within a few hours, but the damage to developer workflows was real. When your coding agent goes down mid-session, you lose context, momentum, and sometimes actual work.

This is the second notable Claude outage in recent weeks, and it tracks with the massive user growth. Claude jumped 40+ spots in the App Store this year. That kind of scaling pressure breaks things. Anthropic needs to invest heavily in infrastructure if they want to keep the users flooding in from the QuitGPT movement.

For API users, the outage mostly affected Claude.ai and Claude Code directly — API endpoints had intermittent issues but stayed more stable. This is actually a good argument for API access over the consumer product: you get more control over retries, fallbacks, and can route through multiple providers.