71% of UK workers admit to using unapproved AI tools at work

29% of organisations have a comprehensive AI governance plan in place

£1.2B in GDPR fines issued in 2025, with AI violations a growing share

‍

For most heritage institutions, public sector bodies, and government-funded organisations, the decision to deploy AI is not simply a technology decision. It is a governance decision. IT security teams must be satisfied that data is protected. Existing policies on acceptable use, cybersecurity, and data handling must be reviewed and updated. And, for organisations where additional investment requires formal approval, trustees and board members must be given a case that is credible, specific, and defensible.

These requirements slow adoption. They are also entirely legitimate. The organisations getting this right in 2026 are not the ones that move fastest. They are the ones that move with the right structure, and arrive at a position where AI is deployed securely, governed transparently, and trusted by the people responsible for oversight.

This article sets out how to navigate each layer of that process: the IT security considerations, the governance and policy framework, and the trustee and board conversation. It is written for the operations lead or internal champion who needs to build a case that survives scrutiny, not just generate interest.

Why Governance Is the Bottleneck, Not the Technology

The technology for AI deployment in operational settings is mature and, for organisations running on Microsoft 365, already present in their existing infrastructure. The bottleneck is not capability. It is the governance process required to deploy it responsibly.

Research from Diligent Institute found that 60% of legal, compliance, and audit leaders now cite technology as their top risk concern, ahead of economic and market risks. Yet only 29% of organisations have a comprehensive AI governance plan in place. That gap between awareness and readiness is where most public sector AI deployments stall.

The November 2025 refresh of the Charity Governance Code introduced a new recommendation that organisations have a formal policy for the use of technology and AI tools, reflecting the growing recognition that trustee-level oversight of AI is not optional. For organisations subject to Charity Commission oversight, this is now a governance expectation. For others operating in regulated environments, the direction of travel is the same.

‍

The Shadow AI Problem Is Already Inside Your Organisation

Organisations that delay approved AI deployment are not preventing AI use. They are removing governance from it. A Deloitte survey of UK workers found that 71% admitted to using unapproved AI tools at work, with 51% doing so at least once a week. IBM's 2025 Cost of a Data Breach Report found that one in five organisations has already experienced a breach linked to unsanctioned AI. The risk of inaction is not zero. It is structural, ongoing, and growing.

‍

Understanding the UK Regulatory Landscape for AI

The UK does not yet have a single dedicated AI Act. Its current framework is principles-based and sector-led, built on five cross-sector principles established in the 2023 government White Paper: safety and security, transparency and explainability, fairness, accountability and governance, and contestability and redress.

In practice, this means AI governance obligations in the UK are distributed across existing regulators. The ICO governs data protection and AI use under UK GDPR. The FCA and PRA apply model risk principles in financial services. For public sector bodies and heritage institutions, the key frameworks to understand are as follows.

UK GDPR and the Data (Use and Access) Act 2025

The Data (Use and Access) Act 2025 received Royal Assent in June 2025, with most data protection provisions coming into force in February 2026. It eases some constraints on automated decision-making while preserving key individual rights. Any AI deployment that processes personal data, including operational data about staff, visitors, or contractors, must comply with UK GDPR requirements around lawful basis, data minimisation, transparency, and subject rights. This is not unique to AI. But AI deployments often involve data flows that were not mapped under previous governance reviews, making a fresh data audit a practical necessity before deployment.

ISO/IEC 42001: The AI Management System Standard

ISO/IEC 42001 is the first internationally recognised standard for AI management systems. It covers risk management, data governance, human oversight, transparency, and accountability across the AI deployment lifecycle. Microsoft has achieved ISO 42001 certification for its Copilot AI products, which provides regulated-sector organisations with a recognised governance baseline when deploying M365-native AI tools. Organisations with existing ISO 27001 certification are, according to practitioners, approximately 40% of the way to ISO 42001 readiness, since the two standards share the same management system structure. For organisations needing to demonstrate AI governance to trustees, regulators, or auditors, ISO 42001 provides the most credible available framework.

The Five UK AI Principles in Practice

The UK government's five principles are not currently backed by statute, but sector regulators are expected to apply them within their own domains. For heritage and public sector organisations, the most practically relevant are accountability and governance (who owns the AI deployment decision, who reviews outputs, who is responsible when something goes wrong) and transparency and explainability (what AI is being used, on what data, and with what human oversight). Both of these translate directly into the documentation and policy work required for trustee sign-off.

‍

What IT Security Teams Need to See Before Approving AI

The IT security review is typically the first formal gate an AI deployment must pass. For organisations with established cybersecurity protocols, this review will centre on four areas.

Data Access and Permissions Mapping

AI tools that access organisational data operate within the permissions structure of the systems they are integrated with. The most common governance failure in early AI deployments, identified by Gartner as affecting over 60% of organisations, is that data permissions were not reviewed before AI was enabled. An AI tool configured with overly broad access can surface sensitive content to users who would not ordinarily have access to it. The remediation is not to avoid AI. It is to conduct a permissions and data classification audit before deployment, applying sensitivity labels to documents and emails and configuring access controls to match the principle of least privilege.

Staying Within the Existing Security Perimeter

The strongest argument for deploying AI within Microsoft 365 in a regulated environment is that it does not require a new security perimeter. Microsoft Entra ID manages access controls. Microsoft Purview handles data classification, audit logging, and compliance. Conditional access policies apply to AI interactions as they apply to any other M365 application. For IT security teams, this is a known and auditable environment. It is fundamentally different from onboarding a new third-party AI vendor, which would require a full vendor security assessment, data processing agreement review, and new integration approvals.

Data Residency and Processing Boundaries

For public sector and heritage organisations handling sensitive operational or personal data, the question of where data is processed and stored is not trivial. UK GDPR requires that personal data transferred outside the UK meets adequacy or equivalent standards. Enterprise AI tools deployed within M365 operate within Microsoft's UK and EU data centres, with data residency commitments documented in Microsoft's data protection agreements. For organisations that cannot use cloud-based AI due to specific contractual or security constraints, on-premises or private cloud deployment options exist but require additional technical architecture consideration.

Audit Trails and Incident Response

IT security approval in regulated environments typically requires evidence that AI interactions can be logged, monitored, and audited. Microsoft 365 provides unified audit logs covering Copilot interactions alongside all other platform activity. These logs integrate with Microsoft Sentinel and other SIEM tools for automated alerting. The ability to demonstrate that AI activity is visible, traceable, and subject to existing incident response procedures is one of the most effective ways to accelerate IT security sign-off.

‍

Pre-Deployment IT Security Checklist

The following steps are typically required before IT security approval: data classification and sensitivity labelling across documents and email; permissions audit to confirm least-privilege access; data residency confirmation for all data processed by the AI tool; vendor security assessment or confirmation of ISO 27001 and ISO 42001 certification; data processing agreement review and signing; and documentation of audit logging and incident response procedures for AI interactions.

‍

Building the Internal Governance Framework

IT security approval addresses the technical risk. Governance addresses the organisational risk. Before a trustee or board conversation takes place, the internal governance framework needs to be in place. This means documented policies, clear accountability, and a structured approach to AI risk that leadership can point to when questions arise.

The AI Acceptable Use Policy

An AI Acceptable Use Policy sets out which tools are approved for use, by whom, on what categories of data, and for what purposes. It also establishes what is prohibited: pasting sensitive data into public AI tools, using personal AI accounts for work purposes, and deploying AI tools outside the approval process. Research consistently shows that providing approved, governed alternatives significantly reduces shadow AI usage. Organisations that have not yet published an acceptable use policy are, in practical terms, leaving their data governance posture undefined.

Roles, Accountability, and the AI Owner

Governance frameworks require named accountability. Someone in the organisation must be responsible for maintaining the AI acceptable use policy, reviewing new AI tool requests, monitoring usage, and escalating issues. In small operations teams, this is typically the Head of Operations or a nominated IT lead rather than a dedicated AI function. The important thing is that the accountability is explicit, documented, and understood by the board.

Risk Assessment and Impact Evaluation

For each AI deployment, a risk assessment should document the intended use case, the data the tool will access, the potential failure modes, the human oversight mechanisms in place, and the review frequency. This does not need to be a lengthy document. It needs to be sufficient for a trustee to satisfy themselves that the deployment has been considered carefully, that risks are understood, and that there is a mechanism for reviewing and correcting the deployment if problems emerge.

‍

Making the Case to Trustees: What Boards Need to Hear

Trustees are not technology evaluators. They are risk stewards. The conversation they need to have about AI is not about what Copilot can do or how Power Automate works. It is about whether the organisation is managing the risks of AI adoption responsibly, and whether the specific deployment being proposed meets the standard of care that trustees are required to apply under charity law or public sector governance obligations.

The November 2025 update to the Charity Governance Code made this explicit: boards should ensure they have the skills, knowledge, and experience to govern effectively in a changing environment. In 2026, that environment explicitly includes AI. Trustees who do not engage with AI governance are not being cautious. They are failing a governance duty.

The trustee case therefore needs to cover five things clearly and concisely.

• What problem this AI deployment solves, expressed in operational terms: hours saved, compliance incidents avoided, consultant spend reduced
• What data will be accessed, how it is classified, and what permissions are in place to limit AI access to appropriate content only
• What governance framework is in place, including the acceptable use policy, named accountability, and risk assessment
• What IT security review has been completed, and what certifications the AI vendor holds (ISO 27001, ISO 42001)
• What the review and escalation process is if problems emerge post-deployment, and how often the deployment will be formally reviewed by leadership

‍

The Framing That Works With Boards

Trustees respond better to operational specificity than to technology vision. The strongest trustee cases do not lead with AI capability. They lead with a specific, quantified operational problem, explain that AI provides a governed solution within existing infrastructure, demonstrate that the risks have been assessed and mitigated, and confirm that the deployment is subject to regular leadership review. That framing turns an uncertain technology conversation into a familiar risk management conversation.

‍

A Phased Approach That Keeps Governance Ahead of Deployment

The most common governance failure in AI deployment is not malice or negligence. It is speed: organisations move into deployment before the governance framework is ready. The phased approach that works best in regulated environments sequences governance work alongside technical work, so that approval gates are met as deployment proceeds rather than after.

1. Phase 1: Phase 1: Governance readiness. Draft the AI acceptable use policy. Conduct the data classification and permissions audit. Identify the named AI accountability lead. Complete the IT security pre-deployment checklist. This phase should take four to six weeks and can proceed in parallel with tool evaluation.

2. Phase 2: Phase 2: Pilot deployment. Deploy to a defined group of five to ten users under IT supervision. Log activity, assess data access patterns, and measure productivity impact against baseline. This phase validates the business case with real evidence before broader rollout.

3. Phase 3: Phase 3: Trustee presentation. Present the risk assessment, governance framework, pilot evidence, and ROI case to the board. This is not a technology demonstration. It is a governance assurance conversation with evidence.

4. Phase 4: Phase 4: Controlled expansion. Roll out to defined user groups with training, an approved tool registry, and clear escalation paths. Update the acceptable use policy as new use cases are identified. Review the deployment formally at six and twelve months.

Why Phasing Matters for Regulated Organisations

A phased approach serves two purposes. First, it produces the evidence that trustees and IT security teams need to give informed approval, rather than asking them to approve something hypothetical. Second, it limits the scope of the initial governance review. Approving a five-person pilot is a materially different conversation from approving organisation-wide deployment. Organisations that start small and demonstrate measurable outcomes move faster through subsequent approval gates than those that attempt enterprise-wide deployment from the outset.

‍

Governance Is Not the Enemy of Speed

The organisations that have deployed AI most successfully in regulated environments are not the ones that bypassed governance. They are the ones that treated governance as the mechanism for building the internal trust that makes deployment sustainable. A deployment that survives its first IT security review and produces evidence that satisfies trustees creates a platform for everything that follows. A deployment that is rolled back after a data incident or blocked by a board resolution sets back the organisation's AI journey by years.

The governance work is real work. But it is finite, and it is the same work that needs to be done regardless of which AI tools are deployed or which use cases are pursued first. Organisations that build the framework now are not just unlocking one deployment. They are building the infrastructure for every AI capability that follows.

‍

Conclusion

VE3 has a proven track record deploying AI within highly regulated public sector and heritage environments. We understand IT security protocols, data governance requirements, and trustee governance structures. If you are preparing an AI business case or navigating sign-off, we can help you build it correctly from the start.

Talk to VE3 about governed AI deployment at ve3.global

‍