📡 THE SIGNAL PREVIEW
The AI company you're switching to just sued the U.S. government.
This week: |
Anthropic's lawsuit — what they actually want
The AI upgrade most of you don't need
The smartest skeptic just raised $1 billion
One prompt if you downloaded Claude this week
📊 THIS WEEK'S POLL
After this week's AI news — what's your next move?
(Results in next week's newsletter 👀)
👤 The Signal
A government document with redacted lines and a red stamp on a dark background
📡 Story 1: Anthropic Filed the Lawsuit — Yesterday
What happened: On March 10, Anthropic officially filed suit challenging the Pentagon's "supply chain risk" designation and Trump's order banning federal agencies from using Claude. The company is drawing its two original lines in legal sand: no mass surveillance, no fully autonomous weapons. Meanwhile, the White House is reportedly drafting new guidelines that would require AI suppliers seeking federal contracts to allow their models to be used for "any lawful purpose" — language that would formalize the same threat Anthropic is fighting in court.
Dan's take: This stopped being a PR stand and became a constitutional challenge. Anthropic is arguing the designation was arbitrary and politically motivated — legally, that's harder for the government to defend than it sounds.
But here's the quieter bomb: that "any lawful purpose" draft order. If it passes, every AI company with ethical red lines faces the same choice Anthropic made. It doesn't just threaten Claude — it sets a precedent for the entire industry.
The market already voted. Since the Pentagon story broke, Claude has hit #1 on the App Store in 16+ countries and is now adding 1M+ new signups daily as of March 6 — confirmed by Anthropic's CPO.
⚡ ⚡ ⚡
A small glowing container outshining a larger empty one — more context isn't always better
📡 Story 2: GPT-5.4 Launched — A 1M-Token Window Sounds Bigger Than It Is
What happened: OpenAI launched GPT-5.4 on March 5, billing it as their "most capable and efficient frontier model." The highlights: a 1M-token context window (roughly 750,000 words — an entire book's worth of text), native computer control for web tasks, and improved step-by-step reasoning. OpenAI also released GPT-5.3 Instant earlier that week. The platform now has 900M weekly active users.
Dan's take: I tested GPT-5.4. The 1M-token window is real. Computer control is beta-level — slow, inconsistent on anything complex. The model is technically better.
But here's the frame that matters: OpenAI is chasing raw power. Claude is building an ecosystem. One million tokens vs. an AI that remembers your client from last Tuesday. That's not the same problem.
Most professionals don't need to process a million tokens — they need continuity. That's a Projects problem. Claude has it. GPT-5.4 has a bigger box.
🔍 If you downloaded Claude this weekend, the full guide is here.
Honest feature breakdown, migration map, and the 5 prompts that actually work — one PDF, everything in one place.
🔥 🔥 🔥
A compass spinning amid broken circuits — the direction of AI is genuinely uncertain
📡 Story 3: The World's Most Prominent AI Skeptic Just Raised $1B to Prove Everyone Else Wrong
What happened: Yann LeCun — Meta's Chief AI Scientist and the most credentialed AI researcher alive — officially launched Advanced Machine Intelligence (AMI) Labs on March 10. The seed round: $1.03B, the largest in European history, at a $3.5B valuation. Backers include Bezos, NVIDIA, Samsung, Eric Schmidt, and Mark Cuban. AMI is headquartered in Paris and building "world models" — AI that reasons about physical reality, not language. (Source)
Dan's take: The skill that compounds isn't knowing Claude or GPT-5.4 — it's learning to work with AI regardless of which tool wins. LeCun just bet $1.03B that the tools everyone's using right now aren't the final answer.
For your workflow, nothing changes this week. But if the smartest person in the room is saying "this whole approach might be wrong," the right response isn't loyalty to any platform — it's staying adaptable.
35% of you already switch between tools depending on the task. That might be the right instinct.
⚡️ SIGNAL VS NOISE
⚡ Anthropic launched the Claude Marketplace — enterprise tools from GitLab, Snowflake, Harvey, and Replit, all built on Claude. Think app store, but for work AI. There's one category of tool already in limited preview that no one is talking about yet. |
⚡ Claude found 22 security vulnerabilities in Firefox in under 20 minutes — including 14 high-severity bugs that human security teams had missed. The Mozilla partnership is public. What Anthropic plans to do with this capability next is not. |
⚡ Microsoft is putting Claude inside Microsoft 365 Copilot — "Copilot Cowork" launches in limited preview late March. The government that blacklisted Claude is about to have it inside every Office suite in America. Nobody in Washington has mentioned this yet. |
⚡ ⚡ ⚡
🔥 Use This Tuesday
The first thing to do if you just downloaded Claude.
21% of you voted that you downloaded Claude after the Pentagon story. Here's your two-minute start: the PDF & Document Digest — Workflow #1 from the cheat sheet.
Copy this prompt. Drop in any PDF. Paste and send.
Summarize this document in 5 bullets.
Pull out the 3 most important numbers or decisions.
Flag anything I should be concerned about.How: Go to claude.ai → click the paperclip → drop in any PDF → paste the prompt above.
That's it. Under two minutes. You'll understand immediately why as of early March, Claude has 11.3M daily active users — up from 4M in January.
The full cheat sheet has all 5 workflows (meeting summaries, spreadsheet analysis, email drafting, content repurposing): [Get the Claude Complete Guide (PDF) →]. Print it or bookmark it.
📊 📊 📊
📊 From The Community
354 of you voted last week . The majority (35%) switch between 2-3 tools depending on the task. But the number I couldn't stop thinking about: 21% downloaded Claude specifically because of the Pentagon story. Another 21% are actively second-guessing ChatGPT. That's 42% of you in some form of flux. Your answers told me what to write this week.
"I use Gemini for brute force or large items, then Claude to fine tune everything. ChatGPT I use for web searches but I'm using that less and less."
"Also will be getting a paid subscription to Claude — I was intending to at some stage with ChatGPT but wouldn't consider it for a second now. Not even a US citizen but if you don't speak up then it'll spread."
"I now honestly believe this recent OpenAI decision has just conceived SkyNet. OpenAI as the mom and DOW the daddy."
👥
I started using Claude because it pushed back on me instead of just telling me what I wanted to hear. Finding out the company behind it walked away from a Pentagon contract for the exact same reason — I didn't see that coming, but it tracks.
Reply and tell me: does the #1 ranking change how you see Claude — or does the missing feature list still hold you back? I'm reading every reply.
Dan Rice · AI Signal · Every Tuesday