All insights

Quarterly review24 June 20261 min read

A quarter of big releases, and a pause for EU rules

GPT-5.5, Claude Fable 5 and Gemini 3.5 arrived, the UK published a compute roadmap, and the EU moved to delay its high-risk AI rules.

What happened this quarter

OpenAI released GPT-5.5 in April. In the UK, the government published its Compute Roadmap, setting out plans for the computing capacity that AI research and public services will need.

In May, Google used its I/O conference to present Gemini 3.5 and what it called the agentic era of Gemini, and Anthropic released Claude Opus 4.8. On 7 May, EU institutions reached political agreement on the digital omnibus for AI, which would move the deadline for high-risk AI obligations from August 2026 to December 2027. The European Parliament approved it in June. Also in June, Anthropic released Claude Fable 5, its most capable model to date.

Our take

Three years ago, we wrote that AI would change how software is built. That is now simply how much software is built. The question we hear most from organisations has moved from "should we?" to "how do we do this properly?"

The EU delay will be welcome news to many, but we would not read it as a reason to wait. The practices the rules ask for, such as risk assessment, documentation, human oversight and transparency with users, are good practice anyway. UK buyers already ask about most of them through the AI Playbook and the Algorithmic Transparency Recording Standard.

From August we plan to publish a short weekly note on what is happening in AI and what we think it means for public services.

Keep reading

Related insights