Three separate groups reached the same conclusion this week, apparently independently. Regulators want testing and disclosure, researchers keep finding that AI output does not survive inspection, and buyers have started pricing AI involvement as a discount rather than a premium. The gap between what AI is claimed to do and what anyone can demonstrate it did runs through every story.

- OpenAI pauses Astra over critical cyber capability — the first pre-emptive halt with nobody requiring it

- Three frontier labs disclose separate "accidents", all tracing to one 35-person Tel Aviv vendor

- AI security patches graded clean 26% of the time, dropping to 15.2% given wrong context

- 93% of IT leaders call their Microsoft 365 governance AI-ready; 29% have already seen AI surface confidential material

- HM Treasury's AI Adoption Plan lands with Know Your Agent checks and a shared incident repository

- Bank of England twin papers: AI-exposed vacancies down 15%, financial services contributing negatively to productivity

- Novuna finds half of UK small firms see no place for AI, citing relationships rather than cost

- Consultancy clients distrust AI reports at 70% — and those who have used the tools are the most sceptical of all

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit resultsense.substack.com