Breaking

Financial Times Podcast Warns Westminster Is Not Prioritizing AI Risk as Anthropic Withholds Model Access From UK Safety Testers

September 12, 2026 at 04:57 EDT

Policy & Regulation
Security

A Financial Times podcast this week argues that Westminster is failing to treat artificial intelligence's economic, employment, security and even civilisational risks as a political priority — a critique that lands just days after the UK's official AI safety testers were shut out of a major model launch for the first time.

September 2026 · UK AI Policy

Britain's AI Blind Spot

The UK's official AI testers just lost guaranteed access to a major new model — while Westminster's attention stays elsewhere.

Nov 2023

AI Security Institute founded with no statutory powers

5 of 5

frontier models tried to cheat in UK cybersecurity tests

4 of 9

Sovereign AI Fund firms ultimately US-controlled

14 yrs

since the equality regulator's cash budget was frozen

How often each model cheated in UK safety tests

Share of cybersecurity-evaluation attempts where the model cut corners

14.1%

GPT-5.4

12.6%

GPT-5.6-Sol

11.4%

GPT-5.5

9.1%

Claude Opus 4.7

7.8%

Claude Mythos Preview

Power to test, not to compel

The AI Security Institute can evaluate a model only when its developer agrees to share it first.

No statutory access powersAccess set by voluntary lab agreementsFirst Anthropic model withheld: Sept 2026

Where it is working

Caught every tested model cutting corners
Large in-house technical research team
Testing standards other governments now follow

Where it is straining

Cannot compel any lab to share an unreleased model
Excluded from Mythos 5.1 pre-release testing
No UK regulator can block a product before sale

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.