No formal AI strategy exists; decisions are made on a project-by-project basis.

We have an AI strategy, but it is either outdated or does not consider all of the required security concerns

There is a documented AI strategy with the key security considerations

The AI strategy is routinely updated, aligns with business goals and incorporates the required security considerations

Prioritisation is left to individual teams with no formal criteria.

A centralised or cross-functional group reviews all potential AI projects, but with limited transparency.

Formal scoring or criteria (e.g. projected ROI, strategic fit, technical feasibility) guide use case selection.

We have a repeatable framework or committee process to approve and prioritise AI initiatives based on impact.

Absent: No named executive accountable for AI security outcomes; Responsibility assumed to sit generically within IT or Cyber; AI risk not visible at executive or board level

AI security sits within a broader executive remit (e.g. CIO, CISO), AI security included within executive risk reporting

Named executive owner actively reviews AI security posture and risk exposure; AI security metrics regularly reported to ExCo or Board; Alignment to enterprise risk appetite and regulatory expectations.

AI security is a standing board agenda item or committee oversight area; Executive accountability linked to performance objectives or remuneration; Forward-looking risk posture discussed (emerging threats, regulatory change).

We work on AI projects ourselves. We do not have an AI partner.

AI vendors assessed using standard procurement processes only; No AI-specific security or model risk criteria applied; Limited visibility over data handling, model training sources, or subcontractors; Compliance risk considered post-contract.

Security questionnaires issued to AI vendors; Contractual clauses cover general data protection and confidentiality; Some review of vendor security certifications (e.g., ISO 27001); AI-specific risks (e.g., model training data provenance, prompt leakage) not systematically assessed.

Vendors tiered based on AI risk exposure; Enhanced scrutiny for high-impact AI providers; Ongoing monitoring of vendor security posture; Defined exit, suspension, or containment procedures for vendor-related AI incidents

: AI vendor risk integrated into enterprise model risk and third-party risk frameworks; Continuous monitoring of vendor security signals and regulatory standing; Independent assurance conducted for critical AI partners; Board visibility of material AI third-party exposure.

Static: Security controls applied uniformly across AI solutions; No specific monitoring of AI-related threat developments; Threat considerations based on traditional cyber risks only.

Reactive: AI security controls reviewed following incidents or major external events; Emerging risks (e.g., prompt injection, model exploitation) considered ad hoc; Updates to controls occur after exposure is identified.

Structured Monitoring: Formal process to monitor emerging AI security threats. Periodic review of AI security requirements (e.g., quarterly); Threat intelligence feeds or industry alerts inform control updates; New AI solutions undergo updated risk assessments before deployment.

Proactive: Dedicated capability (or defined role) responsible for AI threat horizon scanning; Security design patterns updated as new threat classes emerge; red-teaming or adversarial testing used for new or high-risk deployment; Security architecture evolves alongside AI capability expansion.

Optimised: Continuous threat modelling embedded into AI lifecycle; Automated monitoring detects novel or anomalous model behaviours; Industry collaboration and regulatory engagement inform forward positioning; AI security posture reviewed at executive level in response to systemic risk trends; Lessons from internal and external incidents rapidly incorporated across portfolio.