Listen on
Transcript
AI can prepare options, sort information, and suggest next steps. But responsibility does not move to the machine. People still make the decision.
Full transcript of Briefing 07, 0:10, published 5 August 2026.
Key points
- Preparing a decision and making one are different acts
- Accountability follows the person who acts, not the tool that advised
- A recommendation is an input, not an authority
- Governance should name a person for every automated recommendation
The gap between advice and authority
A great deal of AI in professional settings does preparatory work: gathering material, sorting it, ranking options, drafting a recommendation. That is genuinely useful, and it is not the same as deciding. The distinction matters most when something goes wrong, which is exactly when it tends to be least clear.
The practical test is simple. If a recommendation were wrong and acted upon, who would answer for the outcome? If the answer is not a named person, the workflow has a gap in it — and the gap was there before the tool arrived.
Why this gets harder as tools get better
Accountability is easiest to hold when a tool is obviously fallible. As output quality rises, the temptation to treat a recommendation as a conclusion rises with it, and the review step quietly becomes a formality.
Organisations that handle this well tend to write the expectation down: which decisions may be taken on a system's recommendation, which require a documented human review, and what evidence that review leaves behind.
Frequently asked questions
Can accountability for a decision be delegated to an AI system?
No. A system can prepare, rank and recommend, but responsibility for acting on the recommendation stays with the person or organisation that acts. Tools are not accountable parties.
What does good practice look like here?
Naming, in advance, who owns each class of decision, and recording what a human reviewer actually checked. Both matter more than the sophistication of the underlying system.
Does this change as AI systems become more reliable?
It becomes more important, not less. Higher reliability makes review feel unnecessary, which is precisely when a documented review step earns its keep.