Hi Reader!

Two things happening in AI governance this week. We're right in the middle of it.

At 4:30am on Thursday, after six months of negotiation and one trilogue that almost broke the whole package, EU legislators reached a deal to push back the EU AI Act's high-risk obligations. Standalone high-risk systems (biometrics, employment, education, critical infrastructure, justice) move from August 2026 to December 2027. AI embedded in regulated products (medical devices, machinery, toys, lifts) moves to August 2028. Sixteen months of breathing room - or kicking the can down the road, depending on what way you look at it.

This week, Trump will fly to Beijing. He meets President Xi on Thursday, and the Wall Street Journal is reporting that the two sides are exploring a bilateral AI safety dialogue, a possible incident hotline, and some shared language on frontier risk. This is not a treaty, nor a binding framework, but Trump needs a win and this will likely be hailed as the start of a conversation. Trump is annoyed with Europe for not supporting his war on Iran. It plays well for him to build bridges with China in the area of AI regulatory leadership.

The EU has built the only comprehensive AI rulebook on the planet, and last week admitted it cannot enforce it on time.

• The standards aren't ready.
• The national authorities aren't designated.
• The compliance tools companies need to actually meet the obligations don't exist yet.

The decision was made last week to keep the architecture, but let the deadlines slip. This was signalled from as far back as last June. (Check our podcast with Caro Robson for a very interesting chat on China, EU AI Act, and more on governance.) So it’s no big surprise.

The two biggest AI powers are about to talk bilaterally this week, and whatever they agree will be set without Europe in the room. Europe was early to take a stand, back when the message from US was all about ‘unfettered and unregulated access’, but now, the administration is realising that it may need more guardrails than originally thought. Thank Anthropic’s Mythos for that.

What this means for Irish and UK firms

If you were rushing to be ready for August 2026, you now have until December 2027. Use this time to your advantage. The standards are coming, and the firms that build governance frameworks early will move faster when they land.

Anyone we know that is making real ground on AI adoption right now didn't wait around for regulators to tell them what do do. We are living in the age of first movers win fast. Just keep moving.

There's a useful parallel here. When web accessibility was legislated for, it was framed as a cost - something firms had to do for a specific group of users. What actually happened was that accessible websites turned out to load faster, rank better, work on more devices, and were easier for everyone to use. The legislation named one beneficiary but the benefit was universal.

AI governance is shaping up the same way. Data sovereignty, audit trails, human-in-the-loop verification, clear documentation of what your AI does and how - these are the foundations of an AI practice that doesn't leak client IP and doesn't collapse the first time a senior partner asks who signed off on the output. Anyone taking the time now to think their way around how to build the right foundations are doing the right work. Building a version of AI adoption that actually holds up.

This week on ChattingGPT: Joe Toscano

Joe Toscano left Google in 2017 - he's the guy in The Social Dilemma. He now runs Service Stories, helping small firms get found by AI search engines.

Two things: Google just patented the ability to read your website, break it into pieces, and serve searchers an AI-assembled page. They never see your site. If you never invested in SEO, this is your moment to catch up.

And Joe spent two months in China. What he saw — robots, seamless payments, operational fluency — is uncomfortable for anyone who thinks Europe's approach is the only sensible one.

🎧 Apple Podcasts

🎧 Spotify​

Open Cohorts: June and July

Summer at AI Institute

Four public courses run between June and July. They map onto where you actually are with AI right now.

​Browse all public courses >>​

Yours, watching the rulebook and the rule-makers,

AI Institute