TOPIC 05 OF 06
Keep the evidence
Accountability you can showGovernance that only lives in people's heads isn't governance. The durable version is a record: who owns each AI use, what it's allowed to do, how it was tested, what it produced (logs), and what decisions humans made along the way. This is what turns 'we're being responsible' into something you can demonstrate — to leadership, an auditor, or an affected person — and what lets you reconstruct what happened when something goes wrong.
Something looks off in an automated decision. Because there's an owner, a written scope, and logs of what the system did, you can see exactly what happened and correct it — instead of guessing, and instead of discovering there was never any oversight at all.
'A helpful pilot' with no named owner and no records is how un-governed automation slips into production. Decide who owns it and what you'll keep before it touches anything real — retro-fitting accountability after an incident is far harder.
For one AI use, name its accountable owner and the one record you'd most want to have if it went wrong. Start keeping that record.