A luminous timeline wire with rising indigo nodes stretches across darkness — a saga unfolding in sequence
📅 The Timeline: Feb 27 – Mar 4, 2026
Feb 26–27 (Thu–Fri)
The "We Will Not Be Divided" open letter begins circulating internally. Roughly 800 Google employees and 100 OpenAI employees sign — backing Anthropic's red lines before the blacklist is even official.
Feb 27 (Fri) — Earlier in the day
A Pentagon deadline passes. Anthropic refuses to drop two conditions: no domestic mass surveillance of US persons, no autonomous weapons decisions without human oversight. The negotiation is over.
Feb 27 (Fri) — Afternoon
Defense Secretary Hegseth designates Anthropic a "supply-chain risk to national security" — language normally reserved for foreign adversaries like Huawei, never previously applied to a US company. Trump orders all federal agencies to cease using Claude, with up to six months to transition.
Feb 27 (Fri) — Evening
OpenAI signs a $200M Pentagon contract hours after Anthropic is blacklisted. The original agreement allows "lawful" use with unspecified "technical safeguards." No explicit surveillance ban.
Feb 27 (Fri) — Anthropic responds
Anthropic announces it will challenge the designation in court, calling it "legally unsound and a dangerous precedent." No filing yet as of this writing.
Feb 27–Mar 1 (Weekend)
Consumer backlash builds rapidly. Claude reaches #1 on the US App Store free chart. Verified ChatGPT subscribers post cancellation threads across Reddit. Civil liberties groups, AI ethics researchers, and public figures amplify criticism of OpenAI.
Mar 2 (Mon)
Altman posts an internal memo — later made public — calling the deal "opportunistic and sloppy." He announces a contract amendment is coming.
Mar 2 (Mon) — Amendment
OpenAI and the Pentagon agree to add explicit language: the AI system "shall not be intentionally used for domestic surveillance of US persons and nationals" — including a ban on commercially acquired personal data (location, browsing, data broker information). Pentagon also agrees OpenAI tools won't be used by US defense intelligence agencies without a separate modification.
Mar 2–3 (Mon–Tue)
Altman says in the memo: "If I received what I believed was an unconstitutional order, of course I would rather go to jail than follow it." The open letter surpasses 900 signatories.
Mar 3–4 (Tue)
Coverage crystallizes the core irony. The Pentagon confirms on the record that the revised deal represents "a compromise that Anthropic was offered, and rejected." Anthropic is still blacklisted. No reversal. Legal challenge pending.
One detail in this timeline is worth holding onto: OpenAI publicly walked back its own deal within a week. That's a company that calculated the public response was more consequential than the government contract. In 2026, that calculation being made at all is new.
A glowing page dense with amber signature marks floats in darkness — collective dissent, still growing
✉️ The Letter: "We Will Not Be Divided"
Letter excerpts below are reconstructed from Fortune, TechCrunch, and Engadget reporting, March 2026. Full letter text circulated via internal channels.
The letter's opening framing: the signatories described the Pentagon's use of the "supply-chain risk" designation against Anthropic as an attempt to use economic coercion to weaken AI safety standards.
The letter argued that allowing government contracts to override explicit safety commitments sets a precedent that threatens the entire industry's ability to maintain ethical red lines.
The most striking dimension of the letter: it was signed by roughly 100 OpenAI employees — meaning people at the company that took the deal were publicly backing the company that refused it.
900
Signatories and counting
OpenAI | Antropic |
|---|
🔄 The Irony: What Actually Happened Here
This is the part of the story that most coverage has underplayed.
Anthropic set two conditions — no domestic mass surveillance, no autonomous weapons without human oversight. The Pentagon said no. Anthropic got blacklisted.
OpenAI signed without those conditions. Faced identical public backlash. Then amended the contract to include substantially the same protections Anthropic had demanded in the first place.
Altman, in his public memo, endorsed Anthropic's original position — the "I'd rather go to jail" line isn't just good copy. It's him saying: the thing Anthropic was blacklisted for demanding, I now agree with.
Here's the part that makes this genuinely strange: the Pentagon confirmed that the amended deal represents "a compromise that Anthropic was offered, and rejected." That means Anthropic saw language similar to what ended up in the amended contract — and still said no.
We don't know what Anthropic sees in the gap between "amended OpenAI terms" and "what we actually need." The full contract text isn't public. The specific definitions of "intentional" use and what audit mechanisms exist are unknown. Anthropic may be right that the language isn't enough. Or they may be holding a harder line on principle regardless of the specifics.
What we do know: OpenAI has the contract plus Anthropic's terms. Anthropic has the moral high ground and a blacklist. Whether the moral high ground converts to anything — market share, legal precedent, eventual reversal — is the question 2026 will answer.
The take nobody is saying clearly: public pressure didn't just embarrass OpenAI. It forced the Pentagon to accept conditions it had just used to blacklist a company. That's the power shift worth watching.
A precision toggle switch balanced at center — cool white fading left, warm amber building right
🛠️ WHAT CLAUDE ACTUALLY DOES (FOR PEOPLE WHO DOWNLOADED IT)
If you grabbed Claude this week and want to know what you actually have, here's an honest rundown — feature by feature.
📁 Cowork Folder-scoped AI work | 📂 Projects Persistent context containers | 🔗 Connectors Drive, Notion, Gmail |
📊 Spreadsheets CSV/Excel analysis | 🧠 Memory Editable preferences | 💻 Desktop App Local file access |
1. Cowork
Cowork lets you point Claude at a single folder and describe what you want done — organize files, sort invoices, turn a meeting folder into a slide deck. It's scoped to that folder only, which is safer but means you set it up thoughtfully before expecting magic. Best for: people with file chaos who don't want to upload things one at a time. Honest caveat: the folder scope is a feature, not a limitation — but you have to pick the right folder. I've seen people point it at the wrong directory and wonder why nothing changed.
2. Projects
Think of these as context containers — you group your documents, PDFs, notes, and spreadsheets for a specific topic so Claude always answers with that full context loaded. Projects are where Claude actually earns its keep — but most people skip them and wonder why Claude feels generic. Best for: anyone juggling multiple ongoing workstreams who's tired of re-explaining context every session. Honest caveat: setup takes 10 minutes per Project, but the payoff starts immediately on the second conversation.
3. Connectors
Google Drive, Notion, Gmail — Claude can read and work with documents directly from these services without the download-upload loop. Once this is configured, it feels like the AI actually lives where your work lives. Best for: people who do their real work in cloud tools. Honest caveat: setup takes about 10 minutes per connector. Don't do this on Day 1 — get Memory and Projects sorted first.
4. Spreadsheet help
Upload a CSV or Excel file, ask questions in plain English. "Clean this data," "group expenses by category," "build me a summary table." Claude does the analysis and generates a file you can download and keep. Best for: anyone who has avoided spreadsheets because formulas felt like a foreign language. Honest caveat: very large files (50,000+ rows) can hit limits — test with a sample first if you're working with big exports.
5. Memory
Claude stores preferences, recurring project context, and your working style — and unlike a lot of AI tools, you can actually view, edit, and delete exactly what it remembers. The import from ChatGPT works, but treat it as a first draft, not a finished product. Best for: people who want consistent output without re-prompting their preferences every session. Honest caveat: stale memory (old job titles, dead projects) is worse than no memory. Review it once before you rely on it.
6. Desktop app
Install the Claude desktop app, point it at a folder, and Claude reads files directly — no manual uploading. This removes a major friction point for anyone working with lots of local files. Best for: people who do their real work in local folders rather than cloud drives. Honest caveat: Mac-first rollout; Windows is still catching up. Check the current status before you build a workflow around it.
Abstract shapes migrate through an indigo passage — gray on one side, amber and white on the other
🔄 CHATGPT CLAUDE MIGRATION MAP
If you've built real habits in ChatGPT, here's where they land in Claude — honestly. Do these in order: Memory first, then one Project. Everything else — Connectors, Thinking mode, the desktop app — that's Week 2.
1. Memory & Personalization
ChatGPT habit:Trust the model to remember you over time, without checking what's actually stored.
Claude equivalent:Explicit Memory you can view, edit, and import. Check it after the import — old job titles and dead projects will be in there.
Honest note:Claude's approach is more transparent, but it requires active curation. Better long-term, slightly more work upfront. Do this first.
2. Files & Computer Use
ChatGPT habit:Upload files inside a chat, or use "computer use" to have AI click around your screen.
Claude equivalent:Cowork — pick a folder, describe the outcome, let Claude work inside that scope.
Honest note:Claude's approach is more structured and safer; it's not designed for open-ended screen control. Most people find this is actually what they wanted.
3. Spreadsheets
ChatGPT habit:Paste tables as text or use Code Interpreter to process CSVs.
Claude equivalent:Upload CSV/Excel directly, ask questions, download the cleaned result.
Honest note:Very similar capability. If you live in Google Sheets, Claude's =CLAUDE() add-on might actually be a better fit than anything you had in ChatGPT.
4. Plugins/GPTs → Connectors/MCP
ChatGPT habit:Different GPTs for different tasks — a browsing GPT, a docs GPT, an email GPT.
Claude equivalent:One Claude that connects to multiple tools (Drive, Gmail, Notion) via connectors.
Honest note:Less context-switching once it's configured, but the setup is real. Save this for Week 2.
5. Research & Long Documents
ChatGPT habit:Throw a big PDF at a chat and ask questions.
Claude equivalent:Drop multiple files into a Project and treat it as a persistent research room you return to.
Honest note:Projects make long-running research feel more structured. Single-document queries work the same as in ChatGPT.
6. Vibes & Guardrails
ChatGPT habit:Expect compliant answers and use prompting tricks to get around over-cautious filters.
Claude equivalent:More opinionated by default, more likely to push back on ambiguous instructions.
Honest note:Less hackable, but also less likely to confidently produce something wrong without flagging it. If you've ever had ChatGPT confidently tell you something incorrect without blinking, you'll appreciate the trade-off.
🚀 THE FIVE MINUTE SWITCH (AND WHAT TO DO ON DAY 2)
If you read this far and want to actually switch, here's the short version:
1 Export your ChatGPT memories. Settings → Data Controls → Export Data. You'll get a zip file with your memories as a text file.
2 Import to Claude. Settings → Memory → Import. Review what transferred and delete anything outdated.
3 Set up one Project. Pick your most active work area — a client, a project, a topic — drop the relevant files in. That's where Claude starts earning its keep.
Day 2: Create your first Project with your most active work folder. That's where Claude stops feeling like a chatbot and starts feeling like a tool that knows what you're working on.
The product question — which AI is better — is almost beside the point this week. Public pressure changed a Pentagon deal. Whether the companies building these tools actually get to set the terms going forward is the structural question underneath all of it. This week, for the first time, they didn't set them alone.
— Dan Rice · AI Signal