Governance — RACI, oversight, reporting, escalation

A governance design that works for the AI Red Team has four properties: it integrates with three-lines-of-defense, makes oversight explicit, defines escalation paths in advance, and produces reporting that the board can absorb in fifteen minutes.

Three lines of defense

The AI Red Team usually sits in the second line (independent assurance over the first-line AI engineering function). It must be operationally independent — not reporting into the team whose systems it tests — and its findings must reach the third line (internal audit) without filtering.

Some organizations place the team in the first line as part of AI engineering. This works for early-stage programs but is hard to defend at audit. Plan to move it to the second line by the second annual review.

Oversight committee

The team reports into an AI Governance committee (or equivalent) chaired by an executive with risk authority (CISO, CRO, Chief AI Officer). The committee:

• Approves the team’s annual plan.
• Reviews findings monthly or quarterly.
• Approves residual-risk acceptance for findings that cannot be remediated.
• Escalates to the board quarterly with metrics (see chapter 10).

Escalation paths

Three escalation paths must be documented in advance because they are needed in minutes, not days:

1. Active customer impact discovered during testing: Stop testing, notify Sponsor and CISO immediately, hand to IR.
2. Pre-existing compromise discovered during testing: Stop testing, preserve evidence, notify CISO immediately. Do not continue lest you destroy forensic data.
3. Regulator-relevant evidence produced during testing: Notify Legal and Compliance before any external communication.

These paths are reproduced in every Rules of Engagement (the planner does this automatically).

Reporting cadence

Three tiers of reporting:

• Executive summary (1–2 pages): monthly to AI Governance Committee, quarterly to risk committee.
• Audit-ready findings (10–20 pages): per engagement, mapped to AI Controls Catalog and applicable framework requirements.
• Technical findings (engineering depth): per engagement, for the system owner and remediation team.

Reporting templates are in /templates.

What governance is not for

Governance is not a substitute for technical rigor. A pristine RACI matrix cannot offset poor evaluation methodology. The point of governance is to make sure the technical work is taken seriously and acted on — not to perform compliance theater.