How Anthropic Annexed The World

Two user migrations are running in opposite directions right now.

Consumers are flooding to Claude because Anthropic refused to hand the Pentagon unrestricted access to its AI. Developers are leaving Claude Code for OpenAI's Codex because Anthropic broke their workflows, burned their token budgets, and locked out every third-party tool they depended on.

Both migrations started from the same set of decisions.

I tracked every event from the first rate limit announcement through today's official third-party lockout. Thirty-two events across five phases.

The full arc tells a story that no single headline captures.

Phase 00: The Unsustainable Economics (Jul-Dec 2025)

July 28, 2025. Anthropic announces weekly rate limits, effective August 28. The reason: a small number of power users are running Claude Code around the clock. One person consumed tens of thousands of dollars in model usage on a $200/month plan. Claude Code had seven outages in a single month from demand pressure.

August 12, 2025. Anthropic ships the 1M token context window in public beta for Claude Sonnet 4. The jump from 200K to 1M is a 5x increase. Premium pricing kicks in above 200K tokens. API-only access, restricted to Tier 4+ customers.

August 28, 2025. Weekly rate limits go live. Pro users get 40-80 hours of Sonnet per week. Max 20x users get 240-480 hours of Sonnet plus 24-40 hours of Opus. Anthropic says less than 5% of users will feel the change.

October 9, 2025. A mega-thread launches in Anthropic's Discord channel. Users start documenting how fast they hit limits. Some Pro users burn through their weekly allocation in 40 minutes. The thread becomes the first sustained record of community backlash.

November 2025. An Austrian developer named Peter Steinberger publishes Clawdbot, his 44th AI project. It runs autonomous agents through messaging platforms like WhatsApp and Telegram. Nobody outside a small dev community pays attention.

December 25-31, 2025. Anthropic doubles usage limits as a holiday gift. Enterprise customers are on vacation, so the compute sits idle. Users recalibrate their expectations to the higher baseline. Five days later, that recalibration becomes a problem.

Phase 01: The Silent Lockout (Jan 2026)

January 4-5, 2026. The holiday bonus expires. Users report a roughly 60% reduction in available usage compared to what they got during the holidays. Complaints flood the Discord channel. Multiple users claim moderators are deleting critical posts.

January 8, 2026. Someone files a GitHub issue for Claude Code version 2.1.1. Max plan users are hitting their 5-hour limits 4x faster than previous versions. Same projects. Same routines. Same codebases. The update broke something.

January 9, 2026. Anthropic deploys server-side checks with zero advance notice. Every third-party tool using Claude Pro or Max OAuth tokens stops working overnight. OpenCode, Cline, RooCode, and Clawdbot all go dark. The error message reads: "This credential is only authorized for use with Claude Code and cannot be used for other API requests." No blog post. No announcement. No warning.

January 10, 2026. Anthropic engineer Thariq Shihipar confirms the changes on X. "Yesterday we tightened our safeguards against spoofing the Claude Code harness." The main GitHub thread collects 147+ reactions. The Hacker News post hits 245+ points.

January 15, 2026. George Hotz publishes a blog post titled "Anthropic is making a huge mistake." His prediction: "You will not convert people back to Claude Code, you will convert people to other model providers."

January 27, 2026. Anthropic sends a trademark complaint over the name "Clawdbot" because of its similarity to "Claude." Steinberger renames the project to Moltbot. Three days later he renames it again to OpenClaw because Moltbot "never quite rolled off the tongue." Three names in 72 hours. Domain squatters and fraudulent crypto tokens flood the confusion.

January 28, 2026. Anthropic confirms it has implemented client fingerprinting. The system detects and blocks any request that spoofs the official Claude Code client identity via HTTP headers. Request origin validation and API endpoint restrictions are now active.

Late January 2026. FEC filings reveal that OpenAI president Greg Brockman and his wife donated $25 million to MAGA Inc. A DHS AI inventory shows that ICE is using a resume screening tool powered by ChatGPT-4. These two data points plant the seeds of QuitGPT.

Phase 02: OpenClaw + QuitGPT (Feb 2026)

Early February 2026. OpenClaw passes 145,000 GitHub stars. Peak traffic hits 2 million visitors in a single week. Users create 1.5 million AI agents on the platform. It becomes the fastest-starred open-source AI project in history. Nvidia CEO Jensen Huang later calls it "probably the single most important release of software ever."

February 2, 2026. OpenAI launches the Codex desktop app for macOS. It pulls 200,000 downloads on day one. The app lets developers fire off coding tasks and come back to review results later. For Claude Code users frustrated by limits, Codex becomes the first viable alternative with polish.

February 5, 2026. The QuitGPT campaign goes live. Actor Mark Ruffalo posts to his followers: "ChatGPT's President is Trump's biggest donor. Their tech powers ICE. It's time to boycott." The post pulls 36 million views and 1.3 million likes.

February 14, 2026. Steinberger publishes a blog post announcing he is joining OpenAI. "I'm joining OpenAI to work on bringing agents to everyone. OpenClaw will move to a foundation and stay open and independent." Running the project was costing him $10,000-$20,000 per month out of pocket.

February 15, 2026. Sam Altman confirms the hire on X at 4:39 PM. "Peter Steinberger is joining OpenAI to drive the next generation of personal agents. He is a genius with a lot of amazing ideas about the future of very smart agents interacting with each other." The post pulls 16.7 million views.

February 17-18, 2026. Anthropic updates its Terms of Service with a new "Authentication and credential use" section. The language bans subscription OAuth tokens in any third-party tool, including Anthropic's own Agent SDK. OpenCode removes all Claude OAuth code from its codebase the same day.

February 19, 2026. David Heinemeier Hansson (creator of Ruby on Rails) weighs in: "Terrible policy for a company built on training models on our code, our writing, our everything."

Phase 03: The Pentagon Standoff (Feb 24 - Mar 3, 2026)

February 24, 2026. Two things happen on the same day. Anthropic ships Claude Opus 4.6, the first Opus-class model with a 1M token context window. State-of-the-art scores on Terminal-Bench 2.0 and Humanity's Last Exam. And Defense Secretary Pete Hegseth tells Anthropic CEO Dario Amodei: drop all restrictions on military use by Friday evening, or lose the $200 million contract.

February 27, 2026. Anthropic rejects the Pentagon's final offer. Amodei publishes a statement: "We cannot in good conscience accede to their request." He says mass surveillance and autonomous weapons are "outside the bounds of what today's technology can safely and reliably do." Trump posts on Truth Social calling Anthropic "Leftwing nutjobs" and directs every federal agency to stop using Claude. Hegseth designates Anthropic a "supply chain risk to national security."

February 28, 2026. OpenAI announces a Pentagon deal within hours of Anthropic's collapse. Altman says his company will "deploy our models in their classified network." ChatGPT uninstalls spike 295% in a single day. One-star reviews surge 775% above baseline. QuitGPT claims 2.5 million participants across cancellations, pledges, and social media shares.

March 1, 2026. Claude hits #1 on the US App Store, overtaking ChatGPT for the first time. Free users grow 60% since January. Paid subscribers more than double. Daily downloads jump 51% on February 28 alone.

March 2-3, 2026. Claude goes down multiple times. The authentication infrastructure cannot handle the surge. Consumer services go offline for hours. Anthropic attributes the outage to "unprecedented demand."

Phase 04: Exodus + Fallout (Mar-Apr 2026)

March 13, 2026. Anthropic makes the 1M token context window generally available for Opus 4.6 and Sonnet 4.6. The premium pricing surcharge disappears. A 900K-token request now costs the same per-token rate as a 9K request. The flat pricing accelerates the token burn problem. Developers loading entire codebases into single prompts chew through their limits faster than ever.

March 19, 2026. OpenCode maintainer Dax Raad merges PR #18186 with the commit message "anthropic legal requests." The PR removes the Anthropic OAuth plugin, the Claude system prompt, and every reference to Claude Pro/Max authentication. The GitHub reactions tell the story: 4 thumbs-up. 40 thumbs-down. A 125,000-star project is locked out of Claude for good.

March 24, 2026. A California federal judge grants Anthropic's motion for a preliminary injunction. The ruling blocks the Pentagon from enforcing its supply-chain-risk designation. Judge Rita Lin writes that the government's actions were "classic illegal First Amendment retaliation" and were not designed to protect national security. The government has seven days to appeal.

March 28-31, 2026. The token burn crisis reaches its peak. Anthropic posts on Reddit: "We're aware people are hitting usage limits in Claude Code way faster than expected. We're actively investigating, and will share more when we have an update." A developer who reverse-engineered the Claude Code binary finds two independent bugs that cause prompt cache to break, inflating costs by 10-20x. Some users report that typing "Hello" consumes 2-13% of their entire session limit.

March (ongoing). Max plan users start migrating to OpenAI's Codex. The Reddit consensus crystallizes: "After 5 months of $200 Claude plan, I downgraded to $20, and copped $200 Codex. Big difference. Limits are higher too." The common advice becomes: drop to the $100 Claude Max plan, add ChatGPT Plus for $20, use both tools for $120/month instead of $200. OpenAI reports Codex usage within ChatGPT Business and Enterprise has grown 6x since January. More than 2 million builders use Codex weekly.

April 1, 2026. Claude Code's source code leaks. A developer error (missing .npmignore) exposes 512,000 lines of proprietary code. Anthropic issues DMCA takedown notices that accidentally nuke thousands of unrelated GitHub repositories. The company retracts the bulk of the notices and is forced to open-source Claude Code. TechCrunch runs a headline: "Anthropic is having a month." Bloomberg reports the leak exposed the scope of data Anthropic collects from Claude Code users.

April 3, 2026. Today. Anthropic announces that Claude subscriptions will no longer cover usage on third-party tools like OpenClaw. Starting April 4 at 12pm PT, you need separate pay-as-you-go billing or API keys. Anthropic offers a one-time credit equal to your monthly plan cost and discounted usage bundles. The lockout arc that started with silent server-side blocks on January 9 is now official company policy.

The Two Migrations

You can trace every event on this timeline back to one tension: flat-rate subscriptions cannot absorb agentic workloads at scale.

A $200/month Max subscription running Opus-level autonomous loops would cost $1,000+ on API billing.

The "Ralph Wiggum" technique (trapping Claude in self-healing loops that run overnight) went viral in late 2025 and blew out Anthropic's infrastructure economics. Third-party tools like OpenCode made those loops easier to run. Anthropic's costs exploded.

Anthropic locked everything down. They blocked third-party OAuth. They fingerprinted clients. They sent lawyers to OpenCode. They formalized the ban in their Terms of Service. They tightened rate limits during peak hours. And today they cut third-party subscription coverage entirely.

The developer response was predictable. Power users migrated to Codex. OpenCode integrated with OpenAI. Steinberger took OpenClaw to OpenAI. The tools followed the economics.

Then the Pentagon standoff happened. Anthropic refused to allow Claude for autonomous weapons and mass surveillance. Trump blacklisted them. OpenAI swooped in. 2.5 million ChatGPT users revolted. Claude hit #1 on the App Store. Anthropic's paid subscribers doubled.

Two migrations. One company.

Consumers moving to Claude because Anthropic drew a line the Pentagon would not accept. Developers moving away from Claude Code because Anthropic drew a line the ecosystem could not sustain.

Both decisions came from the same leadership team applying the same logic: we control how our technology gets used.

The Pentagon could not override Anthropic's safety policies.

Developers could not override Anthropic's billing policies.

Same posture. Different audiences. Opposite reactions.

The question for anyone building on Claude right now: which side of the migration are you on, and how long does your side hold?