The top 24 AI stories from the last month

Everything that happened in AI in May, in one post

May was the month AI stopped being a productivity story. The Pope wrote 42,000 words about it. China switched on predictive policing. EY had to withdraw a report full of fake citations. SpaceX filed the biggest IPO in history on the promise of a Mars colony. And a New York startup started giving away cleaning services in exchange for footage of the inside of your home.

If you blinked, you missed it. Here are the 24 stories that mattered most.

Peter Thiel takes AI to the ocean

Peter Thiel led a $140 million Series B in Panthalassa, a company building autonomous floating data centres powered by ocean waves. The engineering is clever. The governance, environmental, and accountability questions are concerning.

The AI gap just got institutionalised

Anthropic and OpenAI announced separate multi-billion-dollar consulting arms aimed at enterprise customers, with private equity money behind them. Big businesses get applied engineers and dedicated teams. Small businesses get a £20 subscription and have to work it out on their own. I wrote about this earlier in the year.

AI beats doctors at diagnosis

A study in Science tested OpenAI’s o1 model against ER doctors on 76 real patient cases. AI hit 67% accuracy at triage. The doctors managed 50 to 55%. On long-term treatment plans, AI scored 89%. The doctors scored 34%.

One of the most-cited ChatGPT studies has been retracted

A 2025 meta-analysis claiming ChatGPT improves learning has been retracted by Nature after editors found discrepancies. It had been viewed nearly half a million times and cited in policy. The damage is already done.

Sam Altman’s bad week in court

Three former OpenAI figures testified against Altman in Elon Musk’s lawsuit. Mira Murati called his management “chaotic.” Helen Toner listed the reasons the board fired him in 2023. Under oath, the original concerns came back. In the end, Musk lost the lawsuit. But he did succeed in shining a critical spotlight on OpenAI’s leadership.

ChatGPT can now phone a friend

OpenAI has launched a Trusted Contact feature that warns a nominated friend if it detects signs of self-harm. Three million weekly users are flagged for mental health concerns. The feature is reasonable. The deeper question is why three million people are using ChatGPT for mental health support.

A Chelsea Pensioner was scammed by AI Martin Lewis

Alan Clarke, 72, lost £20,000 of his life savings to an AI deepfake of Martin Lewis on Facebook. Revolut warned him eight times. He went ahead anyway. Social media platforms made £3.8 billion from scam ads in 2025. Should they be cashing in?

Anthropic’s marketing stunt

Anthropic said its new Mythos model was “too dangerous to release” for finding security flaws. The maintainer of curl tested it. One low-severity bug. Three false positives. The PR move underneath is the lesson.

Hey Reader. Do you know what else is dangerously good? My CPD-certified AI Fluency for Leaders course. 3 hours. £225. Find out more here.

A real councillor had to insist he exists

A newly elected Reform councillor in Norfolk was accused online of being AI-generated. The photo was real. His branch admitted using AI on the leaflet, but only to add a countryside background. That was enough to flip the default.

Amazon staff are gaming the AI leaderboard

Amazon set a target for 80% of developers to use AI weekly, then put their usage on a leaderboard. Staff started creating fake tasks to win it. Meta employees are doing the same thing. They call it tokenmaxxing.

Generic phishing is over

Google’s threat intelligence team confirmed that state-sponsored hackers are using AI to map company org charts, identify finance and HR staff, and write phishing emails tailored to individual people. Click-test training is no longer enough.

Anthropic has overtaken OpenAI

For the first time, Anthropic has more business users than OpenAI in Ramp’s monthly AI index. Costs are rising on both sides, customers are leaving for cheaper open-source alternatives, and switching costs are low.

Claude for Small Business

Anthropic has launched a product squarely aimed at the gap I wrote about in my open letter earlier this year. It connects Claude to QuickBooks, PayPal, HubSpot, and a stack of business tools, with a $20 entry price. Credit where it’s due. With major caveats.

The SpaceX IPO is a dream-selling masterclass

SpaceX has filed for the biggest IPO in history. The prospectus tells investors the addressable market is $28.5 trillion. The pay package includes a tranche tied to building a 1 million-strong human colony on Mars.

A real Monet, called slop

The artist SHL0MS posted a Claude Monet painting on X, claimed he’d generated it with AI, and asked users to dissect what was wrong with it. Thousands did. They were certain. They were detailed. They were also wrong.

LinkedIn goes after AI slop

LinkedIn is demoting AI-generated content. The slop wasn’t doing well anyway. The interesting questions are whether they can actually detect it, and whether real writing will finally get the reach it has been earning.

If this story piques your interest, then you probably want to sign up for my upcoming session: LinkedIn for More Leads. LinkedIn can be a minefield or a goldmine.

Ten blue links are over

Google unveiled the biggest overhaul of Search since the search box itself. AI-powered conversational answers, information agents that monitor the web for you, custom widgets generated on the fly. The technology installed as your default route to information is the one most likely to hallucinate.

EY’s AI embarrassment

EY withdrew a cyber security report after researchers found fake data, invented McKinsey citations, and footnotes leading nowhere. The Big Four are not getting this right either. Thoughtful AI use is not a budget question.

Surveillance pricing is here

AI now gives every business the ability to charge each customer the maximum price they will personally pay. Amazon, Delta, Instacart, and JetBlue have all been caught doing something that looks very much like surveillance pricing. Zephyr Teachout has written about it in the FT (it’s a brilliant piece and highly worth a read).

What useful AI actually looks like

Tees Valley Combined Authority used AI to predict traffic congestion and adjust signals automatically. £2 million in. 5,000 hours of waiting saved across a year. Most companies are doing the opposite, with a much worse return.

The Pope on AI

Pope Leo XIV released his first encyclical, 42,000 words entirely about AI. The core argument: technology is never neutral. It takes on the values of those who design, fund, regulate, and deploy it. Your AI choice is a values choice.

China’s predictive policing is live

The Financial Times has documented Chinese authorities rolling out AI surveillance trained to predict crowd build-ups, erratic behaviour, and dissent. Operators can type “a woman wearing a red hat” and retrieve every match. The kit is for sale globally. If this makes you nervous, subscribe to the Humans in the Loop. If it doesn’t make you nervous, definitely subscribe to the Humans in the Loop (I am writing about this next).

OpenAI admits AI breaks the economy

The OpenAI Foundation has committed $250 million to study the economic disruption AI will cause, including research into UBI and taxing capital instead of labour. The proportion is the tell. The PR template is identical to tobacco and oil.

Your home is the product

A New York startup called Shift will clean your apartment for free if you let them film the inside of it. They call the footage training data. The honest word is surveillance. Your data is now valuable enough to swap for physical services.

If you’ve made it this far, you’re in the small minority of people actually keeping up. Most leaders aren’t. That’s not a criticism. It’s just what 24 stories in a month does to a normal calendar.

A few things to take from May:

• The state and the church have weighed in. AI is no longer just a tech story. It’s an ethics story, an economic story, and a values story.

• The gap between large and small businesses is now structural. Both Anthropic and OpenAI have a stance.

• The line between criticism and threat, useful AI and surveillance, training data and exploitation, is being drawn in real time. Your business sits somewhere on that line. Decide where.

If you’d like to go deeper, I run a free monthly Insider’s Briefing covering the month’s stories in 30 minutes. The next one is on the 1st of July. And the next AI Fluency for Leaders course is on the 30th of September (June and July are sold out).

Until July.