Why I cancelled my ChatGPT account

Hello Reader!

I've often compared my AI accounts to children: ask any parent if they have a favourite and the answer (usually) is no. They are all special.

That was my view until very recently.
This bank holiday weekend I unsubscribed from ChatGPT pro - my eldest.

Here's why:

The business case

The way I do my work is using ALL the tools. I jump between them within the one task, picking the strongest tool for the job. Last week I built a deck: did the research and layout on Claude and NotebookLM, graphics on Nanobanana and final slides using Claude projects in our AI Institute template brand and style.

ChatGPT? Nowhere.

Since I discovered Projects on Claude, I'm drawn in. When Cowork launched on Friday, I assumed I'd be glued to Claude even more.

Massive disappointment: not only does Claude Cowork need a mac, it needs one of the newest macs (ie. not with an intel chip), so new shiny toy stayed in the box.

So I caught up reading instead. And the reading gave me another reason to cancel ChatGPT.

The ethical case

I've long admired Anthropic's "race to the top" approach to training its model and I love the fact that they have an in-house interpretability team that regularly shares insights.

In his latest essay "The Adolescence of Technology", Dario Amodei writes chillingly about the risks to humanity if the immense power of AI is mishandled.

"We are considerably closer to real danger in 2026, than 2023", he says.

He goes on to talk about how Anthropic is working to safeguard some of the risks, one of these things is their Constitutional AI - a shared set of values used in post-training. They share so that other technology companies may also use.

Dario Amodei's latest essay, "The Adolescence of Technology", is chilling.
"We are considerably closer to real danger in 2026 than 2023," he writes. He outlines how Anthropic safeguards against risk - including Constitutional AI , a shared set of values used in post-training. this is all shared openly, so other companies can use it too.

I searched for other LLM's that have a constitution and whether they share it. OpenAI? Only moderately transparent. And declining. With all the talk of their profitability struggles, I wouldn't be surprised if safeguarding humanity is slipped down their priorities list.

We're in a new phase.

Agents will proliferate in 2026. Last year was talk. This year, tools we love like Twin—and last week's noise around Clawdot = Moltbot = Open Claw - are showing people what agents can actually do.

We're moving from the chatbot copy-paste era and entering the era of AI assistants that connect to your files and act. I'm excited!

Claude Code is already the developer's choice. Claude Projects has won my heart.

ChatGPT hasn't grabbed me since custom GPTs launched.
When was that again?

The Maths

It's not about cost: I'm paying twice for everything already. We have Team accounts and I have my personal ones. It's more about waste. I was paying for a Pro subscription that I was only really using for meal planning and gardening tips. Meanwhile, the company behind it grows less transparent by the quarter.

My money should go where my values are.

This weekend, I moved everything to Claude.

Saying Goodbye

Before I cancelled, I asked ChatGPT what were the highlights of our time working together. It's answer surprised me:

Credit where it's due: ChatGPT was there at the start. It has helped me think, draft, iterate, and countless times it has brought me joy. It has watched the AI Institute take shape.

But tools evolve, and so do we.
Goodbye, eldest. Thanks for everything.

Jo's compliance advice is surprisingly straightforward. "If what you're doing would give your users the 'ick', there's probably a compliance problem."

Her key takeaway (which we love): stop writing 10,000 word privacy notices and start testing your approach with actual users.

It's so refreshing from a Partner with responsibility for Regulatory, IP, and Digitial at an imminent law firm. Tune in to find out more, including an update on Ireland's rollout of AI regulatory bodies (6th out of 27 EU countries, behind Malta), what qualifies as "high-risk AI," and the reality that junior staff are likely already using ChatGPT with company data.

Many companies are holding off on compliance work, waiting to see if the EU softens the AI Act. Jo's advice is to keep going if you've already started. The relaxation might not happen, and if it does, being ahead on compliance is still the better position.

🎧 Listen on Spotify

🎧 Apple Podcasts

Next Course Starts 12 Feb : AI for Recruiters

For recruiters ready to transform hiring with AI. This is a game-changer for hiring teams.

In this practical, live bootcamp - plus bonus workshop / hackathon - designed especially for Recruiters in conjunction with the Employment Recruitment Federation, you will learn how to use AI with confidence with specific use cases from your daily work.

This is an intermediate level half day workshop for busy recruiters already using AI who want to step up. Don't miss this!