THE AI EDITORIAL — Issue #001
Sunday, March 8, 2026

Anthropic said no to the Pentagon. OpenAI said yes. Here's who actually won.

Hey folks!

Last week, the US government called an American AI company a "supply chain risk."

No, it's not China's DeepSeek or any foreign lab.

Anthropic — the company behind Claude, the model powering hundreds of enterprise stacks, defence contractors, and government systems across the US.

So what happened?

Anthropic had a $200 million contract with the Pentagon. It came with two conditions from their side:

• Don’t use Claude for mass surveillance of American citizens.
• Don’t use it to power autonomous weapons.

The Pentagon said no.

Talks collapsed. President Trump directed federal agencies to stop using Anthropic's tools. Defence Secretary Pete Hegseth officially labelled them a supply chain risk.

And within hours? Sam Altman announced OpenAI had signed the deal.

The same Sam Altman who had publicly said, just days before, that he agreed with Anthropic's redlines. He later admitted the deal was "definitely rushed" and the optics "don't look good.

"OpenAI's own researchers publicly pushed back. One researcher's post saying "I personally don't think this deal was worth it" got 500,000 views. Chalk graffiti criticising OpenAI appeared outside their San Francisco office.

Meanwhile, consumers voted with their thumbs.

Claude hit #1 on the Apple App Store, above ChatGPT, in 20+ countries. Over a million people signed up for Claude daily. Anthropic experienced its largest ever consumer surge, right in the middle of losing a $200M government contract.

What does this actually mean for you?

If you're building on any AI model right now, Claude, GPT, Gemini, this week is a case study in platform risk. One policy call, one negotiation that breaks down, and the tool your entire stack is built on can be reclassified overnight.

This is not a geopolitical story. This is an infrastructure story.

The builders who move fast are the ones who build on AI but never become hostage to any single model.

Fast Break

ChatGPT's market share dropped from 87.2% to 60.4% in exactly one year.

Not because fewer people used it. Weekly users actually doubled from 400 million to 900 million in the same period.

It lost 27 points of market share while growing by 500 million users.

That's what competition at this speed looks like. Growing fast is no longer enough.

Also this week

GPT-5.4 dropped on March 5 OpenAI's new flagship combines reasoning, coding, and agentic workflows into a single model. It can now autonomously control your browser and apps. They also launched ChatGPT inside Excel.

Just imagine: AI that builds your financial models directly in the spreadsheet.

Apple is rebuilding Siri with Google's brain. Apple announced a fully reimagined Siri, powered by Google's Gemini model, that will launch this month alongside iOS 26.4. Yes, the most privacy-obsessed company in tech just handed its AI layer to its biggest rival in mobile.

That's your week in AI — in under 5 minutes.

If something here changed how you're thinking, reply and tell us. We read every response.

See you tomorrow.

The Mextropic AI Team

Building India's AI for the world.