A few months ago, I sat in a meeting where a team proudly presented their new AI usage policy. It had been through legal and two rounds of review. Twenty minutes in, someone mentioned that half the business units were already using AI agents that could act on their own, something the policy never anticipated.
Nobody was careless. The policy was written for a version of AI that had already moved on. Governance runs on committee calendars. Technology runs on release notes.
That creates a growing challenge: how do organizations govern and secure AI when the technology, its use across the business, and the risks it introduces are constantly changing?
Two clocks, two speeds
A governance framework usually takes a year or more to go from idea to something enforceable: draft, comment, approval, roll out, audit. That’s how it should work, because rushed rules are bad rules. Meanwhile, a model that impressed in March may be suspended months later, and yesterday’s demos can quickly become built-in features of tools your staff already use.
Even the big efforts feel this. The EU AI Act took years to negotiate and is being phased in over several more. ISO 42001 and the NIST AI RMF are useful, but they work at the level of principle. They can’t tell you what to do about the plugin your finance team installed last Tuesday.
Why visibility is the real AI governance gap
Most organizations can give you the official AI list: approved vendors and projects that went through intake. They can’t give you the shadow list: free tools people paste data into, AI features switched on by default in software you already pay for, or prototypes that quietly became production.
This is where AI governance becomes a cybersecurity problem. Organizations cannot govern, assess or secure AI use they cannot see.
As AI agents move beyond generating content to taking actions, that visibility becomes even more important. Organizations need to understand who or what is interacting with AI, which data it can access, what permissions it holds, and what actions it is allowed to take.
You can’t assess the risk of something you don’t know exists.
The UAE already has a foundation for AI governance
We often discuss AI governance as if we are starting from a blank page. We aren’t.
The UAE Information Assurance Regulation, issued by the TDRA (formerly the TRA, still known to many as NESA) established the earlier national baseline. The UAE Cyber Security Council subsequently published the UAE Information Assurance (UAE IA) Standard v2.1, dated November 2025, updating that foundation for today’s cyber-risk environment.
The update strengthens rather than changes the central idea: Information Assurance is a lifecycle. Assess risk, apply controls, monitor effectiveness and improve. Version 2.1 retains six management families (M1 to M6) and nine technical families (T1 to T9), while emphasizing technology-agnostic, future-ready, and risk-based implementation. That matters for AI because durable controls should survive the next model, agent or embedded feature.
For UAE organizations, this means AI governance does not have to become an entirely separate security discipline. Existing approaches to risk management, asset management, access control, third-party security, and incident response can provide a practical foundation for managing AI risk.
The question is whether we apply them to AI at all. Too often AI gets its own committee and its own document beside the security program. The policy ends up disconnected from the risk register, asset inventory, vendor process and incident playbook, so nobody can enforce it.
AI governance needs more than a policy
A policy is a document. A capability is the ability to write, apply, check, and change it quickly.
Effective AI governance needs to become an operational security capability, not a periodic compliance exercise.
Mapping AI to the UAE IA Standard v2.1 builds that capability by extending a system you already run:
| UAE IA control | What it means for AI |
| M1 Strategy and planning | Define clear ownership for AI governance and establish the mandate for policies and controls. |
| M2 Risk management | Include AI-specific risks such as data leakage, misuse, unreliable outputs, and unauthorized actions. |
| M3 Awareness and training | Give employees practical guidance on how to use AI securely, including what data should not be shared with AI tools. |
| M6 Performance and improvement | Track AI-related exceptions and incidents, assess control effectiveness, and review regularly. |
| T1 Asset management | Maintain a living inventory of approved AI, embedded AI features, agents, third-party models, and shadow AI. |
| T5 Access control | Define who can use each AI system, what data it can access, and what actions AI agents are authorized to take. |
| T6 Third-party security | Assess AI vendors, third-party models, and embedded AI capabilities as part of third-party risk management. |
| T8 Incident management | Define AI-related incident scenarios, response procedures, and clear ownership for investigation and remediation. |
Asset Management (T1) is where most AI governance conversations end up. Without an AI register, the policy is guesswork, and asking the question in a language the business already understands is how the shadow list comes into the open.
Third-Party Security (T6) closes a gap internal policy can’t. Much of today’s enterprise AI arrives as an update to a product you already license, so ask vendors where data goes, whether it trains their models, and how changes are communicated.
Access Control (T5) also becomes increasingly important as AI evolves from systems that generate answers to agents that can take action. The question is no longer only who can access an AI system. Organizations also need to know which identities the system acts on behalf of, what data it can reach and what it is authorized to do.
One caveat: UAE IA v2.1 is an information assurance baseline, not a complete AI governance framework. Questions such as model bias, explainability and autonomous-agent behaviour still need AI-specific treatment through ISO/IEC 42001, the NIST AI RMF and applicable sector requirements.
The value of UAE IA v2.1 is that it gives those frameworks an operating foundation: ownership, risk management, asset visibility, third-party assurance, incident response and continuous improvement.
What effective AI governance looks like in practice
- Maintain a living AI inventory. Track approved AI, embedded capabilities, third-party models, agents, and shadow AI. Visibility is the starting point for effective AI risk management.
- Govern by outcome, not tool. “No customer data in unapproved AI tools” outlives any product. Controls should be durable enough to survive changes in models, vendors and technology.
- Make the safe path the easy path. If approval takes three months, people won’t wait. Heavy governance doesn’t slow AI down, it pushes it out of sight.
- Keep governance running. Review quarterly, measure real usage, and expect to be slightly behind.
- Invest in people. Awareness beats a 40-page policy nobody reads. Give employees practical guidance that allows them to use AI securely rather than simply telling them what not to do.
- Prepare for failure. AI governance should cover response as well as prevention. Organizations need clear procedures for incidents such as sensitive-data exposure, unauthorized agent activity and compromised AI services.
Keeping AI governance ready for what comes next
Governance will never fully catch up, and it doesn’t need to. A seatbelt doesn’t have to be as fast as the car. It must be there when it matters.
For UAE organizations, the encouraging part is that we don’t start from nothing. The UAE IA Standard v2.1 gives us a lifecycle, a control set and a language that auditors and boards understand. The work is to apply that foundation to AI while complementing it with AI-specific governance and security controls where needed, and keep adapting it as AI changes.
The organizations that do this well won’t have the thickest rulebooks. They will be the ones that can see how AI is being used, understand the risks it introduces and adapt their controls as the technology changes.
Ultimately, keeping pace with AI is not about writing rules faster. It is about building the security and governance capability to keep adapting to what comes next.
References
UAE Cyber Security Council. UAE Information Assurance (UAE IA) Standard v2.1, November 2025. Overview in KPMG UAE, “A bold step towards cyber resilience”: https://assets.kpmg.com/content/dam/kpmgsites/ae/pdf/a-bold-step-towards-cyber-resilience.pdf
UAE Government Portal. UAE Information Assurance Regulation (TDRA, v1.1, March 2020): https://u.ae/en/information-and-services/justice-safety-and-the-law/cyber-safety-and-digital-security/uae-information-assurance-regulation
European Commission, AI Act Service Desk. Timeline for the Implementation of the EU AI Act: https://ai-act-service-desk.ec.europa.eu/en/ai-act/timeline/timeline-implementation-eu-ai-act
ISO/IEC 42001:2023, Information technology — Artificial intelligence — Management system: https://www.iso.org/standard/42001
NIST. AI Risk Management Framework (AI RMF 1.0): https://www.nist.gov/itl/ai-risk-management-framework
