Risk Acceptance Memo Template for Buying AI Controls From an Early-Stage Vendor
A defensible risk acceptance memo states the decision and scope, lists each standard criterion the vendor does not meet, names the vendor-specific risks of an in-path control, records the compensating controls and the residual risk that remains, assigns a named business owner, and sets review triggers, an expiry date and exit conditions before approvers sign.
Check before sign-off
- Every unmet criterion has a stated response option and rationale.
- Failure behavior of the in-path control is stated and evidenced by the vendor.
- Each compensating control has a scope and a named owner.
- Residual risk is written out and disclosed to the teams relying on the control.
- One business-side owner has acknowledged the risk in writing.
- Expiry date, review triggers and exit conditions are set.
Section 4: compensating controls, residual risk and risk transfer
Deployment model
Self-managed deployment in your own cloud reduces reliance on vendor-operated infrastructure. Record who operates it and who patches it.
Data residency and retention
Record the configured settings and the owner who can change them.
Evidence portability
Confirm evidence can be exported or retained in storage you control, in a format your auditors can read.
Contractual protections
Where negotiated, document transition assistance, escrow or similar arrangements and what they cover.
Exit and transition plan
Name the alternative control or fallback path, the owner and the date it was last exercised.
What the memo has to decide
A risk acceptance memo records a decision to proceed with a purchase that does not clear a standard threshold. The NIST AI RMF Playbook lists four organizational responses to AI risk: avoid, accept, mitigate or transfer. The memo should state which response applies to each gap rather than treating the whole purchase as a single yes or no.
An early-stage vendor commonly misses criteria such as operating history, financial durability and concentration risk. When approval is granted informally, nothing records the residual risk, the compensating controls, the owner or the conditions for revisiting the decision. The Playbook defines residual risk as the sum of all unmitigated risks and says it should be documented, with accepted risks denoted in the risk response plan.
A control in the AI execution path raises the stakes. If it fails, AI availability and the continuity of audit evidence can both be affected. The memo should therefore be time-bound and evidence-backed, not a procurement formality.
Sections 1 and 2: decision, scope and unmet criteria
Section 1 states the decision in one sentence. Suggested wording: "Approve [vendor] to provide [control function] for [systems, models, agents, business units] under an exception to [policy reference], for a period ending [date]." Then define scope: which environments and data classes are covered, which are excluded, and what the control is allowed to do to an interaction.
Section 2 lists every standard third-party criterion the vendor does not meet. For each, record the response option (accept, mitigate, transfer or avoid) and the reason. Record the risk tolerance that applies and why this acquisition falls inside or outside it. If the purchase deviates from policy, cite the policy and the deviation explicitly.
One published supply chain case study illustrates the logic. A security exception protocol required formal acceptance of risk from the business unit leader whenever the business wanted more vendor risk than the vendor's rating would justify. The exception was a documented approval, not an informal one.
Section 3: vendor-specific risks of an in-path control
Generic vendor risk language is not enough here. Write each of the following as a specific statement with the evidence reviewed.
Failure behavior. State whether the control fails open or fails closed when it is unavailable. Fail-open keeps AI workloads running but lets interactions proceed without enforcement. Fail-closed preserves enforcement but interrupts the business process. The memo should name the chosen behavior, who chose it and what the vendor has demonstrated. Ask how often the vendor performs resilience testing.
Change control. Ask how vendor updates are released and what reassessment happens before they reach your environment. Make an update to the in-path control a reassessment trigger.
Support and key-person dependency. Identify who can respond to an incident, how escalation works and whether your team can operate the control if specific vendor staff become unavailable.
Financial viability and continuity. Record what financial evidence you reviewed and what happens to the deployment if the vendor is acquired, restructured or stops operating.
Evidence continuity. State whether you can still show what the control allowed, redacted or blocked for the full retention period if the vendor relationship ends.
Sections 5 to 7: ownership, review triggers, expiry and approval
Section 5 names one accountable owner on the business side. That person acknowledges the risk and accepts responsibility for adverse impact from the deviation. Security and risk functions advise and challenge, but they should not be the default owner. Also record who is responsible for ongoing management of the control, and how residual risk is disclosed to internal teams that rely on it.
Section 6 sets review triggers. Useful ones include a vendor update to the in-path control, a change in vendor ownership or financial position, an incident or failure event, a change in scope, a lapse in any compensating control, and a re-calibration of your risk tolerance. Set a regular review cadence by policy, since published guidance calls for regular review but gives no fixed interval.
Section 7 sets an expiry date after which the acceptance lapses unless renewed, and defines exit conditions that start the transition plan. Close with the approver block: business owner, risk, security, legal or procurement, and any authority your exception policy requires, each with date and signature.
Memo structure at a glance
Label: 1. Decision and scope
Description: What is being approved, for which systems, under which policy exception.
Label: 2. Unmet criteria
Description: Each standard threshold missed and the response option chosen.
Label: 3. In-path risks
Description: Failure behavior, updates, support dependency, viability, evidence continuity.
Label: 4. Compensating controls
Description: Each control with its scope and owner, plus residual risk.
Label: 5. Ownership and approval
Description: Named accountable owner and required approvers.
Label: 6. Review and exit
Description: Triggers, expiry date and conditions that start transition.
Test the memo against a real deployment
Talk through self-managed deployment, retention settings and evidence export for your own memo.
Talk to an ExpertHow does your AI governance program compare?
See where your program has gaps in less than 2 minutes.