AI & MCP risk and controls matrix
Source: deck slides 6–9 · Wrike AI & MCP Readiness Programme
Each risk name links to the consolidated risk view, which sets the Wrike vendor team's response from the WPP × Wrike Pilot deck beside the programme position.
Columns on every table: AI and MCP show whether the risk applies to AI agents and to MCP (Y = yes, N = no). Required For is the stage by which the control is needed (Pilot, Rollout or GA).
Governance, data, identity, MCP security and accuracy#
Slide 6 · Risks #1–#5
| Risk Area | Key Risk | Potential Impact | Proposed Controls | Wrike Proposed Capability | Global Wrike Team Controls | AI | MCP | Accountable Owner | Required For |
|---|---|---|---|---|---|---|---|---|---|
| #1 Governance & Accountability | Agents created without review or ownership. | Uncontrolled deployment; unclear accountability. | Registration, approval, governance review and named owners. | Agent Registry; Approval Workflow; Named Ownership. | Manual Taskforce review and approval before activation; central register; named owner; time-limited approval. Secondary owner, automated owner alerts and mature retirement workflow can follow at Rollout. | Y | Y | Wrike Product Team + Global Wrike Team | Pilot |
| #2 Data Exposure & Scale | Programmatic retrieval, extraction or update at scale. | Data leakage, mass updates or mass data corruption. | User-scoped access, monitoring and approval controls. | Scoped Service Accounts; High-Impact Action Approval. | Delete disabled; approved spaces, fields and actions only; bulk update/extraction limits; record-return and rate limits; time-, scope- and volume-bounded approval; preview for high-impact actions. | Y | Y | Wrike Product Team | Pilot |
| #3 Permissions & Identity | Excessive permissions or privileged accounts. | Broader access and weakened traceability. | Least privilege and recurring recertification. | Scoped MCP Tokens; No Shared Credentials. | Least privilege; no shared credentials; approved identity; manual access check before activation. Automated recertification reporting can follow at Rollout. | Y | Y | Wrike Product Team + Global Wrike Team | Pilot |
| #4 MCP Security | Weak authentication or token misuse. | Data or system compromise. | OAuth, token management and restricted access. | OAuth; IP Restrictions; Controlled Token Issuance. | OAuth or approved authentication; controlled token issuance; expiry, revocation and rotation; no shared token. Central token inventory/reporting can mature at Rollout. | N | Y | Wrike Product Team + Global Wrike Team | Pilot |
| #5 Accuracy & Hallucination | Incorrect, incomplete or misleading responses. | Wrong decisions and reduced trust. | Validation and human review. | Not currently identified. | Pilot outputs must be manually validated by the Taskforce and users; no unreviewed reliance on generated output. Formal source attribution, confidence indicators and scaled validation workflow required by Rollout. | Y | Y | Wrike Product Team + Global Wrike Team | Rollout |
Process, monitoring, consumption, support and compliance#
Slide 7 · Risks #6–#10
| Risk Area | Key Risk | Potential Impact | Proposed Controls | Wrike Proposed Capability | Global Wrike Team Controls | AI | MCP | Accountable Owner | Required For |
|---|---|---|---|---|---|---|---|---|---|
| #6 Process & Approval Bypass | AI circumvents established controls. | Governance breach and unauthorised outcomes. | Human approval and audit review. | Agent Approval Workflow. | Taskforce approval of each use case plus separate approval for bulk, workflow, permission or high-impact actions. Agent cannot alter access controls or override approval rules. | Y | Y | Wrike Product Team | Pilot |
| #7 Monitoring & Visibility | No visibility of agents, owners or actions. | Weak oversight and investigations. | Registry, dashboards and scheduled reviews. | Agent Registry; Dashboard; Activity Feed. | Manual register and review of all five Pilot agents and actions. Enterprise dashboard, automated ownership reporting and anomaly monitoring required by Rollout. | Y | Y | Wrike Product Team | Rollout |
| #8 Token Consumption & Cost | Usage grows without ownership. | Budget overruns and allocation challenges. | Quotas and consumption monitoring. | Usage Quotas; Cost Attribution. | Manual usage and cost review is acceptable for Pilot/Rollout while volumes are controlled. Quotas, forecasting and cost attribution required before GA. | Y | Y | Wrike Product Team + Global Wrike Team | GA |
| #9 Operational Support | No clear support ownership. | Longer incidents and support burden. | Support, escalation, disablement and recovery. | Support Model; Central Management. | Named support contact, incident route, recovery procedure and individual-agent disablement required for Pilot. Formal SLA and scaled support model can mature later. | Y | Y | Wrike Product Team + Global Wrike Team | Pilot |
| #10 Compliance & Regulatory | Regulatory or contractual conflicts. | Compliance exposure. | Governance reviews and approvals. | Audit Logging. | Taskforce screens every Pilot use case against client, market, contractual and data restrictions; retain approval evidence. Full compliance reporting and retention model required before GA. | Y | Y | Global Wrike Team + Business Taskforce | Pilot |
Sprawl, assurance, vendor dependency and reputation#
Slide 8 · Risks #11–#14
| Risk Area | Key Risk | Potential Impact | Proposed Controls | Wrike Proposed Capability | Global Wrike Team Controls | AI | MCP | Accountable Owner | Required For |
|---|---|---|---|---|---|---|---|---|---|
| #11 Agent Sprawl | Duplicate and unmanaged agents. | Complexity and support burden. | Registry, standards and rationalisation. | Agent Registry; Naming Standards. | Manual register is sufficient for five Pilot agents. Naming standards, approved templates, duplicate review and retirement workflow required by Rollout. | Y | N | Global Wrike Team | Rollout |
| #12 Audit & Assurance | Activity cannot be evidenced. | Audit findings and failed assurance reviews. | Audit trail and recurring reviews. | Immutable Audit Logging. | Retain Pilot approval, owner, configuration/version and activity evidence; identify user versus agent. Exportable records and formal retention/assurance model can mature before GA. | Y | Y | Wrike Product Team | Pilot |
| #13 Vendor Dependency | Required enterprise controls are unavailable. | Governance gaps remain open. | Gap assessment, roadmap follow-up and compensating controls. | Product Commitment. | Track product gaps from Pilot and use compensating controls. Formal roadmap commitments and delivery dates are required before GA where unresolved gaps affect scale. | Y | Y | Wrike Product Team | GA |
| #14 Reputation & Client Trust | AI incidents reduce trust. | Adoption resistance and client concerns. | Controlled rollout and oversight. | Not identified. | Taskforce-approved use cases, client-impact review, named communications owner and incident route required for Pilot. | Y | Y | Global Wrike Team | Pilot |
Tool use, data boundaries, identity, prompt injection, traceability and autonomous action#
Slide 9 · Risks #15–#21
| Risk Area | Key Risk | Potential Impact | Proposed Controls | Wrike Proposed Capability | Global Wrike Team Controls | AI | MCP | Accountable Owner | Required For |
|---|---|---|---|---|---|---|---|---|---|
| #15 Tool Misuse | Agent performs unintended actions. | Operational disruption or data impact. | Restrict tools to approved use cases. | Scoped MCP Tokens. | Delete disabled; approved tool catalogue; restrict write tools to approved use case; individual-agent disablement. Finer tool-level administration can mature at Rollout. | N | Y | Wrike Product Team | Pilot |
| #16 Cross-System Data Leakage | Information exposed across connected systems. | Client confidentiality and contractual risks. | Data boundaries and classification controls. | Scoped MCP Tokens. | Pilot limited to approved connections, spaces and data; restrict context and returned volume; no sensitive cross-system flow unless specifically approved. | N | Y | Wrike Product Team + AI CoE + Global Wrike Team | Pilot |
| #17 Identity Propagation | Identity misuse across systems. | Loss of accountability. | User-scoped authentication. | Scoped Service Accounts. | User-scoped identity; no shared accounts; delegated access checked before activation; approval expires with identity or access change. | N | Y | Wrike Product Team + Global Wrike Team | Pilot |
| #18 Prompt Injection | Malicious instructions alter AI behaviour. | Unauthorised extraction, updates or actions. | Restricted tools, monitoring and training. | Not confirmed. | Manual use-case review is not sufficient alone: restrict tools, limit bulk actions and require approval for high-impact writes. Advanced prompt-injection detection can follow at Rollout. | Y | Y | Wrike Product Team | Pilot |
| #19 Lack of Traceability | Requester or action cannot be identified. | Difficult investigation and root-cause analysis. | End-to-end audit logging. | Immutable Audit Logging. | Pilot must identify requester, agent and resulting action. MCP connection detail, export and longer retention can mature at Rollout/GA. | Y | Y | Wrike Product Team | Pilot |
| #20 Autonomous Actions | High-impact actions execute without oversight. | Mass data corruption or operational disruption. | Human approval gates. | Agent Approval Workflow. | Delete disabled; update limits, bounded approvals, high-impact preview, individual kill switch and manual review. Rollback, protected fields and automated anomaly detection can mature at Rollout. | Y | Y | Wrike Product Team | Pilot |
| #21 External AI Platform | Data leaves approved governance boundary. | Compliance and governance risk. | Approved architecture and review process. | Registry; Approval Workflow. | External platform/connection architecture must be approved before Pilot use; register each connection and restrict scope. Enterprise-standard hosting/gateway maturity required before GA. | N | Y | AI CoE + Global Wrike Team | Pilot |