The Role of NEDs in Overseeing AI-Powered Businesses
The Short Answer
An AI-powered business is one where artificial intelligence is the product, the service or the engine of the business model: a company that sells a model, licenses an AI platform, or delivers a service that only works because a model makes the decisions. For a company like this, the questions a board usually treats as technical are commercial and legal ones. If the model fails, the product fails. If the training data was not properly licensed, the core asset may be compromised. If a regulator classes the use as high risk, the route to market changes.
The non-executive director’s job is not to understand the mathematics. It is to make sure the board knows what the model does, where its data comes from, who owns what, how it is tested, what happens when it goes wrong, and how all of that affects the company’s value. That requires enough AI literacy to ask sharp questions and the independence to keep asking until the answers make sense.
This article is about companies whose business is AI. If you want the broader question of how any board oversees AI use, strategy and ethics, read our companion piece on whether NEDs are ready to oversee AI, which covers the general framework.
Why AI-Native Companies Are a Different Oversight Problem
Most boards now oversee some use of AI, usually in operations, marketing or customer service. In those businesses AI is one risk among many, and a failure is serious but survivable. In an AI-native company the position is reversed. The model is the asset, the source of revenue and the main source of risk at the same time.
Three features make the oversight task harder.
The core asset is hard to see. A manufacturing board can walk the factory floor. An AI company’s main asset is a model trained on data, refined through testing and deployed through software. Its quality, limits and failure modes are not visible in the board pack unless someone deliberately makes them visible.
The business can change faster than the board meets. A model can be retrained, a provider can change its terms, a new open-source release can undercut the product, and a regulator can publish guidance, all between two quarterly meetings. Boards that rely only on scheduled papers will be behind.
The founders and engineers usually know far more than the board. That information gap exists on every board, but it is wider here. Non-executives have to work harder to get information that is complete, balanced and understandable, and to avoid deferring to technical authority simply because it is technical.
None of this changes the legal position. Every director, executive or not, owes the same duties under the Companies Act 2006, including the duty to exercise reasonable care, skill and diligence. Our guide to the legal duties of a non-executive director explains what that standard means in practice. In an AI company, meeting it means engaging with the model, not just the financials.
Model Risk: The Product Itself
Model risk is the risk that the model gives wrong, biased, unstable or unexplainable results, and that those results cause harm to customers, the company or third parties. In an AI-native company it is the equivalent of product quality risk, and it deserves the same seriousness.
A board does not need to audit the model. It does need to know that a credible testing and monitoring regime exists, and to see evidence that it works. Useful questions include:
- What is the model designed to do, and what is it explicitly not designed to do?
- How is performance measured, against what benchmark, and how has it changed over time?
- How do we detect drift, where performance degrades as real-world data moves away from the training data?
- What testing happens before a new version goes live, and who signs it off?
- How do we test for bias and unfair outcomes, especially where the model affects people’s access to services, credit, jobs or prices?
- Can we explain an individual decision to a customer or regulator if asked?
- What are the known failure modes, and what happens to customers when the model fails?
The answers should be in the board pack in a form non-specialists can follow: a short model risk dashboard with performance, incidents, drift indicators and testing status, reported every meeting. If management cannot produce that, the absence is itself the finding. Our article on appointing an AI ethics board member covers where specialist expertise helps.
Data Rights: Where the Model Came From
An AI model is only as defensible as the data it was trained on. For an AI-native company, the provenance of training data is a balance-sheet issue as much as a compliance one.
Non-executive directors should understand, at a high level, the sources of the company’s training and operating data and the rights the company holds over each. The main categories are usually:
- Customer data, used under contract terms and privacy notices that may or may not allow model training.
- Licensed data, bought or licensed from third parties under terms that may restrict use, territory or duration.
- Publicly available data, which is not the same as freely usable data. Copyright, database rights, website terms and data protection law can all still apply.
- Synthetic or generated data, whose quality and lineage need their own controls.
Where personal data is involved, UK GDPR applies throughout the model’s life cycle: collection, training, testing and deployment. The ICO’s guidance on AI and data protection is the starting point, and a board of an AI company should expect the data protection officer or equivalent to report on how the company meets it, including lawful basis, transparency, data minimisation and individuals’ rights.
The board question is simple to ask and hard to answer: if a major customer, investor or regulator asked us to prove we had the right to use every material dataset behind our model, could we? A business heading for investment or sale will be asked exactly that in due diligence.
Intellectual Property: What the Company Actually Owns
Investors value an AI company on what it owns and can defend. That is often less clear than founders assume. The board should understand the company’s position on four questions.
Who owns the model? Where a model is built on a third-party foundation model or open-source components, the company owns its own fine-tuning, data and integration work, but the base is licensed. The licence terms, including any restrictions on commercial use, matter.
Who owns the outputs? UK copyright law has a specific provision for computer-generated works, in section 9(3) of the Copyright, Designs and Patents Act 1988, but how it applies to modern generative AI is unsettled. Customer contracts should be clear about who owns what the product generates and what warranties the company gives.
Are we exposed to infringement claims? If training data included copyright material without permission, or outputs reproduce protected works, the company may face claims. The board should know whether this has been assessed and what indemnities the company gives to customers or receives from suppliers.
Is our know-how protected? Much of an AI company’s edge is in its methods, data pipelines and tuning. Confidentiality, employment contracts and access controls protect it. Staff turnover in AI is high, so this is a live risk.
These are legal questions that need legal advice, and a NED should make sure the company is getting it rather than try to answer them personally.
Safety, Security and Resilience
AI products introduce failure modes that conventional software does not. A model can be manipulated through the data it receives, can leak information it was trained on, or can produce harmful outputs that were never intended. When the product is the model, these are product safety and cyber security issues combined.
The board should expect management to cover, at least annually and after any incident:
- how the product is tested against misuse and adversarial inputs before release;
- what guardrails limit harmful or inappropriate outputs, and how they are monitored;
- how the model, its weights and training data are secured against theft or tampering;
- how the company depends on cloud and model providers, and what happens if a key provider changes its terms, pricing or availability;
- what the incident response plan is for an AI failure, including who decides to pull or roll back a model, and how customers and regulators are told.
Concentration on a single foundation-model or cloud provider is a particular risk for many AI companies. A change in a provider’s pricing or terms can move margins overnight. A board should see that dependency mapped and understand the fallback options. Our article on why boards want NEDs with cyber risk expertise covers the security side, and our guide to the NED’s role in a cyber crisis covers incident response.
Regulatory Exposure
The UK has taken a principles-based approach to AI, asking existing regulators to apply cross-cutting principles within their sectors rather than creating a single AI law. The government’s AI regulation white paper sets out that approach. In practice, an AI-native company’s UK regulatory exposure depends on what the product does and who uses it: data protection through the ICO, consumer protection and competition through the CMA, and sector regulators such as the FCA for financial services or the MHRA for medical devices.
Companies that sell into the EU face more prescriptive rules. The EU AI Act applies to providers placing AI systems on the EU market and can reach UK companies whose AI outputs are used in the EU. Banned practices have applied since February 2025 and obligations for general-purpose AI models since August 2025. The EU’s AI omnibus regulation, which came into force in July 2026, pushed back the high-risk deadlines: obligations for stand-alone high-risk systems now apply from 2 December 2027, and for AI built into regulated products from 2 August 2028.
For an AI-native company this is strategic, not just compliance. Whether a product is classed as high risk affects development cost, time to market, the documentation customers will demand and the company’s valuation. The board should know which of its products, current and planned, could fall into the higher-risk categories, and how the roadmap reflects that. The extra time is a planning window, not a reason to wait.
For a regulated AI business, for example one providing services to FCA-authorised firms or itself authorised, the Senior Managers and Certification Regime adds personal accountability. Our guide to FCA-regulated board governance explains how that works at board level, and our page on fintech NED recruitment covers the boards where AI and regulation meet most often.
Valuation and Investor Expectations
AI-native companies are often valued on expectations rather than current profits, and those expectations are sensitive to the issues above. The board’s oversight of model, data and IP risk is therefore also oversight of the company’s value.
Non-executive directors add particular value in three areas.
Keeping the narrative honest. AI companies face strong pressure to describe their products in the most ambitious terms. Overstating what AI can do is a legal risk, through consumer protection and advertising rules and, for regulated firms, financial promotion rules, as well as a reputational one. The board should be comfortable that public claims match what the product actually does.
Unit economics. The cost of compute, data and model providers can make an AI product look attractive at small scale and uneconomic at large scale. NEDs should push for clear reporting of gross margin per customer or per query, and how it changes as usage grows.
Due diligence readiness. Investors and acquirers will test data rights, IP ownership, model performance and regulatory classification. A board that has been asking these questions all along will be ready. Our articles on valuations and investor expectations and building a board that attracts investors cover the wider picture.
How the Board Should Organise Itself
There is no single right structure, but AI-native boards that oversee well tend to share some features.
AI risk is a standing board item, not an occasional deep dive. Because the model is the business, model risk, data and regulatory status should appear in every board pack.
There is a clear executive owner. Someone below board level, often a chief technology officer, chief AI officer or head of responsible AI, is accountable for model governance and reports to the board on it.
Committees have clear remits. Many AI companies extend the audit and risk committee to cover model risk. Larger ones create a technology or responsible AI committee. What matters is that someone on the board owns each risk and that nothing falls between committees. Our guide to board committees and where NEDs add most value explains the options.
The board has at least one director with relevant technical depth. This does not mean every NED must be a technologist. It means the board as a whole should be able to test management’s technical claims. A tech-literate independent director, combined with others who bring commercial, legal, regulatory and financial experience, usually works better than a board of engineers. Our pages on technology NED recruitment and AI transformation expertise describe the profiles boards look for.
The board builds its own literacy. Regular briefings, demonstrations of the product, and sessions with the people who build and test it help non-specialist directors ask better questions. Our guide to NED induction and development covers how boards organise this.
Who Owns What: AI Oversight on the Board
| Who | Responsibility |
|---|---|
| The board | Sets the appetite for AI risk and approves the strategy |
| Audit or risk committee | Oversees model risk, data controls and assurance |
| Remuneration committee | Considers whether incentives encourage safe use of AI |
| An executive owner | Accountable for AI use and its risks across the business |
| A technology-literate NED | Helps the board ask informed questions; does not run the technology |
Jodie says
AI expertise on boards is still scarce
Boards ask us for a NED with AI experience, but the candidates who have overseen AI at board level are still few. We often suggest a director with strong data and technology risk experience, supported by an expert adviser to the board.
That combination usually works better than waiting for the perfect candidate.
Jodie Garrington, Partner, NED Capital
Questions Every NED Should Ask an AI-Native Board
If you join, or are considering joining, the board of a company whose business is AI, these questions are a good starting point:
- What exactly does our model do, what does it not do, and how do we know?
- Can we prove our rights to every material dataset used to train and run it?
- What do we own outright, what do we license, and what restrictions come with the licences?
- What are our top three model failure modes, and what would each mean for customers?
- How dependent are we on a single model or cloud provider, and what is our fallback?
- Which of our products could be classed as high risk under the EU AI Act or a UK sector regulator?
- Do our public and sales claims match what the product does?
- How do our margins change as usage grows?
- Who is accountable for model governance, and when did they last report to the board?
- Would our data, IP and testing records stand up to investor due diligence tomorrow?
A board that can answer these clearly is governing the business it actually runs. One that cannot has found its agenda for the next few meetings. For wider thinking on longer-term AI risk, see our article on whether boards should prepare for AGI risks.
Frequently Asked Questions
What is an AI-powered or AI-native business?
It is a business where AI is the product, the service or the engine of the business model, rather than a tool used in support functions. Examples include companies that sell or license models, build AI platforms, or deliver services that depend on a model’s decisions. For these businesses, model, data and regulatory risks are core commercial risks.
Does a NED on an AI company board need a technical background?
Not every NED. The board as a whole needs enough technical depth to test management’s claims, which often means one director with relevant technology experience. Other directors add value through commercial, legal, regulatory and financial experience, provided they build enough AI literacy to ask good questions.
What should be in the board pack for an AI-native company?
A regular, concise view of model performance, incidents, drift, testing status, data and IP issues, provider dependencies and regulatory developments, alongside the usual financial and operational reporting. It should be written so non-specialists can follow it.
Does the EU AI Act apply to UK AI companies?
It can, if a UK company places AI systems on the EU market or its AI outputs are used in the EU. Banned practices and general-purpose AI rules already apply. High-risk obligations have been delayed to 2 December 2027 for stand-alone systems and 2 August 2028 for AI built into regulated products.
Who is responsible if an AI product causes harm?
The company is responsible for its products, and directors remain responsible for overseeing the company under the Companies Act. Depending on the product and sector, consumer protection, data protection and sector-specific rules may also apply. Boards should take legal advice on their own exposure.
How do we find a NED with AI experience?
Start by defining the gap: technical depth, regulatory experience, commercial scaling, or experience of investment and exit in AI businesses. Then search beyond the founders’ networks. Our tech NED recruitment service runs searches for exactly these roles.
Related Guides and Services
Oversight, expertise and recruitment for technology and AI boards.
AI Oversight
Governing AI
→ Are NEDs Ready to Oversee AI?
→ AI Ethics Board Member
→ AGI Risks for Boards
Recruitment
Technology NEDs
→ Tech NED Recruitment
→ Technology Sector NEDs
→ AI Transformation Expertise
Board Practice
Risk and Value
→ Cyber Risk Expertise
→ Valuations and Investors
→ Board Governance Hub
Adrian Lawrence FCA — Founder, NED Capital
Fellow of the ICAEW | Chartered Accountant holding an ICAEW practising certificate | Sister practice of FD Capital
Adrian holds a BSc from Queen Mary College, University of London and is a former listed-company Finance Director with more than 25 years at C-suite level, working with boards, investors and business owners across the UK. He founded NED Capital to connect organisations with the independent non-executive directors they need, and oversees the senior search team that runs every board mandate we accept.
Need a NED Who Understands AI?
Call 0203 137 2496 or email recruitment@nedcapital.co.uk to discuss a non-executive or chair appointment for an AI or technology business. NED Capital is part of FD Capital Group, five specialist recruitment firms, with offices at 167–169 Great Portland Street, London W1W 5PF and Watling Court, Orbital Plaza, Cannock WS11 0EL.
NED Capital | Part of FD Capital Group | Founded by Adrian Lawrence FCA, ICAEW practising certificate holder
Adrian Lawrence FCA is the founder of NED Capital and a Fellow of the Institute of Chartered Accountants in England and Wales (ICAEW) and holds an ICAEW practising certificate in his own name. He holds a BSc from Queen Mary College, University of London, and has more than 25 years at C-suite level, working with boards, investors and business owners across the UK. He founded NED Capital to connect businesses with the independent Non-Executive Directors they need to provide challenge, governance and strategic oversight — and personally leads candidate assessments for board-level appointments.