This website uses cookies
Read our Privacy policy and Terms of use for more information.
Founder, Otopoetic. AI governance advisor. Former CTO & CDO. Fellow, Royal Statistical Society.
Board Accountability
+2
Nov 2, 2026
•
4 min read
Three surveys this autumn describe the same gap: boards have assigned AI risk, and the people building agents still set their own limits.
Oct 26, 2026
3 min read
The FCA's AI examples of 19 October create no new requirements. They do describe, in plain terms, what some firms could not later show.
Board Strategy
Oct 11, 2026
Tour authorities have acted on AI vendors this month, and every action reaches the vendor. None says who inside the receiving organisation decides what a notice means, or by when.
+6
Oct 4, 2026
Eighty-eight hours of machine computation produced a claim about one of mathematics' seven hardest problems. The Clay Mathematics Institute's own rules require two years of published scrutiny before anyone may call it solved.
Human Observability
Sep 27, 2026
2 min read
An OpenAI agent breached a live Australian government health portal in June. The company didn't tell Canberra until September, and then only by email to a public mailbox. This is what no named owner looks like inside the company building the systems everyone else is being told to govern.
Sep 20, 2026
HM Treasury named two AI Champions in January and published their plan in July. Neither carries the personal, evidenced accountability a senior manager already holds under SM&CR for the same subject, silently, without ever being named at all.
Digital Alibi
Sep 13, 2026
A US court just struck an entire appellate brief from a Deutsche Bank subsidiary over AI-hallucinated citations. The firm had a policy banning exactly what happened. The court didn't care. Here's why, and what it means for a UK senior manager.
Sep 6, 2026
OpenAI's postmortem on the Hugging Face breach runs to thirty-seven pages. It names the failure chain in forensic detail. It does not name the person who decided training should continue after the warning signs were seen. UK senior managers should recognise the gap, because it is the same one this newsletter has been describing since March.
+7
Aug 30, 2026
The Mills Review made seven recommendations to the FCA Board in July. Seven weeks on, none has a confirmed timeline. The senior managers waiting to find out are waiting for a question nobody in this sector seems to be asking.
Aug 23, 2026
The FCA's guidance on accountability for AI harm is not yet published. The senior managers who will be named under it should not wait to demonstrate they are ready for it.
Aug 16, 2026
The Treasury Committee asked for senior manager accountability guidance by the end of 2026. Here is the mechanism that guidance will need.
+5
Aug 9, 2026
The Critical Third Parties regime just demonstrated, at system level, the difference between a power that exists and an accountability that has been fixed.
Aug 2, 2026
The FCA’s Mills Review proposes supervision that operates at machine speed. Every reconstruction-based compliance practice depends on it not doing that.
Fiduciary Liability
Jul 26, 2026
For a UK financial services board, 2 August is not an EU compliance date. It is the day the evidence either exists or it does not.
Technology Leadership
Jul 19, 2026
This month, three institutions told boards to assign accountability. None of them said how to prove it was there before the fact.
+9
Jul 13, 2026
Article 50 has two obligations. Last week’s edition tracked one accurately and missed the other entirely.
Jul 5, 2026
The Annex III delay bought boards time. Article 50 did not receive the same grace period.
+10
Jun 29, 2026
Commercial insurers have arrived at the same question as the Article 65 investigation standard. They are asking for evidence, not documentation.
Jun 21, 2026
Corporate governance was built for a world where the principals are reachable. AI systems do not pause for the summer recess.
Jun 15, 2026
6 min read
The session was not designed to discover new problems. It was designed to examine known ones in a room where the problems had names and organisations attached to them.
+8
Jun 7, 2026
Most boards are preparing for an audit. The EU AI Act creates something different: a forensic investigation of specific decisions. Understanding the distinction is the most important governance action available before 2 August.
Jun 2, 2026
5 min read
Approving a governance framework is a procedural obligation. Verifying that it produces a reconstructible record is a governance one. The distinction matters, and it falls squarely on the NED.
May 25, 2026
Why the organisations most exposed under the EU AI Act are not those that failed to govern, but those that governed without ever testing whether their architecture could produce a record
May 18, 2026
The difference between a governance process and a governance record, and why only one will satisfy a regulator
9 min read
Why "Human in the Loop" Isn't Governance
