I’ve been recommending ChatGPT (owned by OpenAI) to clients for a couple of years. It’s powerful, it’s recognisable, and most people have at least heard of it. But over the past year or so I’ve been quietly shifting my own work over to Claude, made by a company called Anthropic, and I think it’s time I explained why.

This isn’t a features comparison post. There are plenty of those. This is about something a bit less exciting but actually more important: who’s running the show, and whether you can trust them with your business.

A bit of background

Anthropic was founded in 2021 by a group of people who left OpenAI. Not because of money or ambition. Because they thought OpenAI was moving in the wrong direction.

Dario and Daniela Amodei, along with several colleagues, had growing concerns about governance, the increasing pull of commercial pressure, and what they saw as a drift away from safety as a genuine priority. So they left and built something new, with safety explicitly at the centre of what they do.

That’s not something made up for the press, it’s genuinely their origin story.

What’s been happening at OpenAI

You might have caught some of the drama. In late 2023, the board fired Sam Altman, then reinstated him within days. What most coverage glossed over was that the underlying tension was about safety culture and accountability, not just internal politics.

Since then: OpenAI quietly removed the word “safely” from its mission statement during a restructuring. Its dedicated safety team was disbanded in early 2024. Multiple senior safety researchers resigned. Until public pressure forced a change, staff who left were reportedly bound by non-disparagement agreements that prevented them from speaking out about internal safety concerns.

None of this means ChatGPT is dangerous to use. But it does paint a picture of a company where safety went from being the point to being a marketing line.

What Anthropic does differently

Anthropic publishes something called a Constitution, an actual written document that sets out what Claude is supposed to value and why. It’s not hidden in a policy PDF nobody reads. It’s public, it gets updated, and it explains the reasoning rather than just issuing a list of rules.

The priority order they’ve set out is: safety first, then ethics, then Anthropic’s guidelines, then helpfulness. They’re explicit that helpfulness comes last, not because it doesn’t matter, but because a model that’s helpful but unsafe or dishonest isn’t actually useful to anyone.

They also acknowledge openly that AI models can be wrong or confused, which is why human oversight needs to stay in place. That’s a remarkably candid thing for an AI company to say about its own product.

Then the Pentagon got involved

If you want a very clear illustration of the difference between these two companies, look at what happened recently with the US Department of Defense.

The Pentagon’s AI strategy pushed for frontier models that could be used for “all lawful purposes,” a phrase that, when you unpack it, includes mass domestic surveillance and fully autonomous weapons systems with no meaningful human oversight. Anthropic, whose Claude models were already running on classified networks, said no. Hard no. No mass surveillance of Americans, no autonomous weapons without a human in the loop.

The Pentagon reportedly floated using emergency powers to force the issue. Anthropic still said no. The contract, worth around $200 million, collapsed. Anthropic was subsequently labelled a “supply chain risk” by the US government.

Most companies, when faced with losing a $200 million contract and being branded a national security problem, would find a way to say yes.

Within hours of the US government ordering agencies to stop using Anthropic, OpenAI announced its own Pentagon deal to supply AI for classified military networks. This wasn’t entirely surprising: OpenAI had quietly removed its blanket ban on military and warfare uses in 2024, a policy shift that received relatively little attention at the time but now looks rather significant in hindsight.

Sam Altman later described the initial agreement as “opportunistic and sloppy” and amended it after staff and public backlash. ChatGPT uninstalls reportedly spiked by nearly 300%. Staff resigned. The hashtag CancelGPT trended.

OpenAI walked through the door Anthropic refused to open. That contrast is not subtle.

Why this matters if you’re a small business owner

You might be thinking: I just want to write emails and summarise meeting notes, I’m not building autonomous weapons. Fair enough. But if you’re using AI in your business, you’re building habits and workflows around a particular tool, and you’re choosing which companies to support.

I’d rather point people towards a tool built by people who walked away from $200 million rather than compromise on their stated values, than one whose creators quietly rewrote their ethics policy and then called it opportunistic when it backfired.

Anthropic is the smaller company. It has less market share, less name recognition, and apparently less interest in making friends with the Pentagon. In the AI world, that currently makes it the small but mighty option.

I’m not saying they’re perfect. No company is. But they’ve at least made their position clear, and when someone tried to change it and hand them a very large cheque, they said no.

That’s a small but mighty difference.

If you want to give Claude a try, I’m happy to help you get started in a Power Hour session. We can look at your current AI tools and work out what’s actually serving you.