Is our picture of AI risk current — and does it reach the people who own the consequences?
Enhanced: 22 controls
Who it applies to
Enter for every AI system whose letter is E or A (this area has no Baseline controls). Go as deep as your system letter, and implement in your mode.
Why it exists
Keep the risk picture current as the world moves, and carry it to the level that can act on it — because AI risk changes faster than any periodic review: new attacks emerge, models change, uses expand, regulations shift. A risk assessment is accurate on the day it is written and decays from there. Watch for what changes, refresh the picture, and report it through the governance bodies, so that the people who own the consequences see the risk while they can still act on it.
What to actually do
Track the changing risk landscape and refresh the picture. Watch for new threats, changed models, expanded uses, and shifting rules, and update the risk catalogue and assessments as they move — driven by the change events the earlier domains surface, including the provider model-change warning.
Report risk upward through the governance bodies, at proportionate depth. Feed the current risk picture into the reporting path of the Governance oversight dimension — committee, risk function, board — so AI risk is governed alongside every other enterprise risk, not in a silo of its own.
Controls
Enhanced2 controls
Control 229
Refresh the risk catalogue and assessments as the landscape changes.
EnhancedDetectiveall rolesper system
What it is
Driven by change events, not the calendar.
How to implement it · Minimum rules
Triggers: provider model-change notice, new model version, new use or user group, new data source or connected tool
Also triggers: a newly publicised attack technique, a change in applicable rules, an incident or near miss
On a trigger: reassess the affected risks before the change goes live, or within 30 days if it is external
Each trigger feed has an owner: e.g. IT for provider notices, legal or management for rule changes
Log every refresh: trigger, date, risks reviewed, changes made
How to start
List the change events and who will tell the risk owner about each; subscribe to every AI provider's change notices.
Evidence
The refresh log linking each change event to the updates made in the catalogue and assessments.
Mode: light / standard / full
light
The AI lead receives provider notices by email and updates the risk spreadsheet.
standard
Change management asks 'does this affect AI risk?' and routes yes answers to the risk lead.
full
Change events open review tasks in the risk tool automatically.
Control 230
Report AI risk upward through the governance bodies.
EnhancedDetectiveall rolesper system
What it is
At proportionate depth, feeding enterprise risk management.
How to implement it · Minimum structure
Top AI risks with current rating and trend since the last report
Risks above appetite and the status of their treatment
Residual risks accepted since the last report, and by whom
New risks and incidents arising from change events
Decisions requested from the body receiving the report
Cadence: e.g. quarterly to the committee or management, at least annually to the board; immediately for risks above appetite
How to start
Add an AI risk item with the top five risks to the agenda of the next management or risk meeting.
Evidence
The reports and minutes, and the AI risks entered in the enterprise risk register.
Mode: light / standard / full
light
The AI lead gives management a one-page summary at a regular meeting.
standard
The risk lead reports quarterly to the AI committee and feeds the enterprise risk report.
full
Risk integrates AI into reporting to the risk committee and board, with trends.
Evidence
The risk-refresh history against change events; the risk reporting into the governance bodies; evidence AI risk is integrated with enterprise risk.