Risk register: where each risk is covered
The deck refers to risks by number (#1 to #21) but never lists them in one place. This register is assembled from the slides. The "Concern" column quotes the deck, with the slide in brackets. Where one risk has several concern wordings, the slides use it from different angles.
Page key: Own = Governance: ownership and visibility · Review = Governance: review and action oversight · Identity = Identity and permissions · Conn = MCP connections · Scope = Data scope · Human = Human review · Audit = Audit and traceability · Tools = Tools and data boundaries · Incident = Disablement and incident response
| # | Risk | Concern, as worded in the deck | Covered on |
|---|---|---|---|
| 1 | Governance & Accountability | Agents lack review or ownership. (s3) | Own |
| 2 | Data Exposure & Scale | Large-scale retrieval or updates. (s10) Bulk actions affect unintended records. (s12) | Scope, Human |
| 3 | Permissions & Identity | Excess access; unintended actions. (s8, with #15) Unapproved applications gain access. (s9) Privilege changes or rule bypass. (s10, with #6) | Identity, Conn, Scope |
| 4 | MCP Security | Weak authentication or token misuse. (s8) An integration connects to the wrong account. (s9) Previously issued tokens retain access. (s9) | Identity, Conn |
| 5 | Not in the deck | See discussion points | |
| 6 | Process & Approval Bypass | Established controls are bypassed. (s5) Privilege changes or rule bypass. (s10, with #3) AI bypasses established controls. (s12) | Review, Scope, Human |
| 7 | Monitoring & Visibility | Agents, owners or actions are unseen. (s3) | Own |
| 8 | Not in the deck | See discussion points | |
| 9 | Operational Support | Unclear ownership prolongs incidents. (s17) | Incident |
| 10 | Compliance & Regulatory | Contractual or regulatory conflicts. (s14) | Audit |
| 11 | Agent Sprawl | Duplicate or unmanaged agents. (s3) | Own |
| 12 | Audit & Assurance | Activity cannot be evidenced. (s14) | Audit |
| 13 | Not in the deck | See discussion points | |
| 14 | Reputation & Client Trust | AI incidents reduce trust. (s17) | Incident |
| 15 | Tool Misuse | Excess access; unintended actions. (s8, with #3) Unintended actions disrupt work. (s16) | Identity, Tools |
| 16 | Cross-System Data Leakage | Data crosses approved boundaries. (s9, s16, with #21) | Conn, Tools |
| 17 | Identity Propagation | Identity misuse across systems. (s8) | Identity |
| 18 | Prompt Injection | Malicious instructions alter behaviour. (s16) | Tools |
| 19 | Lack of Traceability | Requester or action is unidentified. (s14) | Audit |
| 20 | Autonomous Actions | High-impact actions lack oversight. (s5, s12) High-impact changes at scale. (s10) Damaging actions continue unchecked. (s17) | Review, Scope, Human, Incident |
| 21 | External AI Platform | Data leaves approved boundaries. (s5) Data crosses approved boundaries. (s9, s16, with #16) | Review, Conn, Tools |
Risks #5, #8 and #13 do not appear anywhere in the deck. The rows are kept so the gaps are visible. The Pilot readiness page groups the risks that are covered into three sets:
- Governance & oversight: #1, #6, #7, #11, #20
- Identity, data & tool boundaries: #2, #3, #4, #15–#18, #21
- Audit, support & client trust: #9, #10, #12, #14, #19
Discussion log#
| Date | Topic | Decision / outcome | Owner |
|---|---|---|---|