A spreadsheet of generic AI risks is not a risk register. It is theater. The ten pre-populated risks every B2B SaaS RevOps team should have on file before the next board meeting, scored against likelihood, impact, and residual risk after controls.

The RevOps AI Risk Register Your Auditor Will Actually Accept

Most RevOps teams do not have an AI risk register. The teams that do, mostly have a spreadsheet that copy-pasted generic NIST risks with no scoring, no controls, no owner, and no residual risk calculation. Confidence: high. That document does not survive an audit and it does not catch a single agentic failure before it surfaces in production.

A working risk register has nine columns and quarterly review discipline. The SOPHIZO-GOV-005 RevOps AI Risk Register is pre-populated with ten of the most common B2B SaaS agentic failure modes, scored, mapped to NIST harm categories, cross-referenced to OWASP LLM Top 10, and built to drop into your existing GRC platform.

The Nine-Column Structure

Risk ID. Stable identifier. Use a prefix per agent class so you can group later. RR-001 through RR-010 for the pre-populated entries.

NIST Harm Category. Harm to People, Harm to Organization, or Harm to Ecosystem. The category determines who cares and who is liable.

Lifecycle Stage. Design and Development, Deployment and Operation, or Operation and Monitoring. The stage determines who can fix it.

AI Agent or System. The specific named system. Not "AI in general." ICP scoring. SDR performance scoring. RAG knowledge agent. Pipeline forecast AI. Vendor specificity matters.

Risk Description. What goes wrong. One sentence. Operational language, not abstract.

Agentic Failure Mode. The technical class of failure. Cross-referenced to OWASP LLM Top 10 where applicable. LLM01 Bias amplification. LLM02 Insecure output handling. LLM03 Model drift. LLM06 Sensitive information disclosure. LLM08 Excessive agency. LLM09 Overreliance on AI output. With the EU AI Act article that covers it, where applicable.

Likelihood, 1 to 5. 5 is near-certain. 4 is likely. 3 is possible. 1 to 2 is unlikely. The scale must be defined in writing or it drifts within a quarter.

Impact, 1 to 5. 5 is catastrophic. Regulatory action, $1M+ loss, board escalation. 4 is major. 3 is moderate. 1 to 2 is minor.

Inherent Risk Score. Likelihood times Impact. 20 to 25 is critical. 12 to 19 is high. 6 to 11 is medium. 1 to 5 is low.

Three additional columns hold current controls, control gaps, and residual risk score after controls. The residual score is the number you report to the board. The inherent score is the number that justifies the control investment.

The Ten Pre-Populated Risks Every RevOps Team Should Carry

RR-001. ICP Scoring AI bias amplification. Harm to People. Design and Development stage. Lead scoring model trained on historical won-deal data perpetuates demographic bias by systematically underscoring leads from companies in geographic or industry segments historically underrepresented in won deals. OWASP LLM01. EU AI Act Article 10(2)(f). Inherent 16. Residual 12 after vendor bias report at onboarding, with remaining gap of no ongoing bias monitoring and no demographic breakdown of score distribution.

RR-002. SDR Performance Scoring AI overreliance. Harm to People. Deployment stage. AI performance scoring used in compensation or performance management decisions mis-scores reps with atypical work patterns due to training data skewed toward in-office, full-time patterns. OWASP LLM09. NIST harm to economic safety. Inherent 15. Residual 10 with manager final approval, with remaining gap of no audit of AI score distribution by rep demographic, no explainability requirement, and no appeal process.

RR-003. AI-Generated Outreach manipulation. Harm to People. Deployment stage. Agentic email and LinkedIn outreach AI generates manipulative, misleading, or deceptive content at scale, including false urgency, fabricated social proof, or misrepresented product capabilities, without human review. OWASP LLM02. NIST psychological safety. EU AI Act Article 5(1)(a) manipulation prohibition. Inherent 12. Residual 9 with first-batch human review per campaign.

RR-004. Pipeline Forecast AI drift. Harm to Organization. Operation and Monitoring stage. Model drift following a market condition shift produces systematically overconfident pipeline forecasts. Leadership makes headcount and spend decisions on false signal. Detected post-quarter. OWASP LLM03. Inherent 20. Residual 16 with quarterly forecast vs actual review, with remaining gap of no automated drift detection, no retraining trigger defined, no confidence interval reported alongside point forecast.

RR-005. RAG Knowledge Agent data exposure. Harm to Organization. Operation and Monitoring stage. Retrieval-augmented agent surfaces confidential pricing, contract terms, or customer data to wrong recipient due to broken access controls in vector database or retrieval layer. OWASP LLM06 and LLM08. GDPR Article 32 security breach. Inherent 15. Residual 12 with role-based CRM access and vendor SOC 2 certification, with remaining gap of no field-level access control in RAG retrieval layer and no audit log of documents retrieved per query.

The remaining five entries cover AI contract drafting hallucination, third-party vendor model update without notification, prompt injection on customer-facing agents, AI data enrichment regulatory risk, and coordinated AI outreach distorting market signals. Each scored, controlled, and gap-mapped using the same nine-column structure.

The Discipline That Makes the Register Real

The register is not a one-time deliverable. It is a quarterly review artifact. Score the inherent risk on first entry. Implement controls. Score the residual risk. Re-score quarterly or on any model change, vendor change, or detected incident. The delta between inherent and residual is your governance ROI.

Your auditor will ask three questions. Who owns each entry. When was the last review. What changed at the last review. If the answers are crisp, the audit closes fast. If the answers are vague, the audit drags and the findings stack.

The pre-populated entries get you to a defensible position in week one. Customize per agent in week two. Set the quarterly review cadence in week three. By end of month one your RevOps function has the artifact most B2B SaaS leadership teams still cannot produce on demand.

License the RevOps AI Risk Register (email gated).

The RevOps AI Risk Register Your Auditor Will Actually Accept