Board Briefing - AI, Cyber Risk and Proportionate Oversight

With the rapid growth of AI, how worried should Boards be about cyber risk, and should they be asking for different or additional assurance?

That question reflects the feeling that AI is being positioned as both an opportunity and a threat. Boards are rightly asking how to support innovation while remaining confident that resilience and governance are not being compromised.

Artificial intelligence is reshaping how organisations operate, it presents significant opportunity

  • Improved productivity

  • Faster analysis and decision support

  • Service optimisation

  • Better targeting of resources

  • Automation of repetitive processes

At the same time, AI is changing the cyber and digital risk landscape.

It increases the speed and sophistication of attacks, expands dependency risk across supply chains, and introduces new data and model governance considerations.

The Board’s role is to ensure that opportunity is realised within clear guardrails, and that governance keeps pace without becoming a barrier.

Effective oversight is about balance. Too little governance creates unmanaged exposure, too much creates unnecessary friction. The focus should be proportionate control, visibility and accountability.


1. AI as a Cyber Threat Accelerator

AI is being used by attackers to improve phishing, impersonation, vulnerability discovery and automated reconnaissance.

These are familiar risks, but AI has increased the tempo. Where controls are weak, exposure happens faster.

Where fundamentals are strong, AI does not invalidate them.

Board oversight should focus on whether core controls are consistently applied and properly assured.


2. AI as a Data Risk Accelerator

AI systems are heavily dependent on data. They consume, generate and transform large volumes of information at speed.

As AI adoption increases, data risk accelerates in parallel.

This includes:

  • Greater volume of personal data processing

  • Expanded data sharing across platforms

  • Increased reliance on third-party cloud services

  • More complex data lineage and transformation

  • Automated generation of derived data, classifications and risk scores

AI systems are only as reliable as the data they rely on.

Where data quality or integrity is poor, AI does not correct the problem.

It amplifies it.

Inaccurate, incomplete or biased data can result in flawed outputs being produced at scale and at speed.

AI is a multiplier of both data value and data weakness.

 

Risks can arise where:

  • Data is reused beyond its original purpose

  • Sensitive data is incorporated into AI workflows without appropriate controls

  • Data retention and minimisation disciplines are diluted

  • Outputs are stored without clear provenance

  • Data lineage becomes opaque

  • Data transfers occur through embedded supplier AI features

 

Boards should expect assurance that:

  • Data flows involving AI are documented

  • Data Protection Impact Assessments are completed where required

  • Data lineage and provenance are understood

  • Data quality controls are applied to both input and generated data

  • Retention policies extend to AI-generated outputs

  • Supplier data processing involving AI is visible and governed

 

AI does not remove existing data protection and governance obligations. It increases the speed and scale at which data is processed and decisions are influenced.

If your organisation is still working on improving data quality and data governance, then you should think carefully about whether you are ready to layer additional complexity or whether you should get the foundations right first.


3. AI-Specific Data and Model Risks

AI also introduces new governance considerations that sit across cyber, data protection and operational risk.

Synthetic and Generated Data

AI systems may generate scores, summaries or classifications that influence decisions. The logic and results used must be explainable, verifiable, and able to be interpreted correctly.

Risk arises where generated outputs are treated as verified data, stored without traceability, or incorporated into tenant records without audit clarity.

Boards should expect assurance that data quality governance extends to AI-generated outputs.

Model Drift

Over time, models can degrade, behave inconsistently or introduce bias.

Initial approval is not sufficient, ongoing monitoring and validation is required.

Boards should expect visibility of how higher-impact AI systems are reviewed and revalidated to ensure data and logic can be trusted.

Prompt and Output Risk

Where generative AI or chatbots are deployed, risk can arise through both inputs and outputs.

On the input side:

  • Sensitive information may be entered into external AI systems without appropriate controls

  • Staff may inadvertently disclose personal or confidential data in prompts

  • External users may attempt to manipulate the system through adversarial prompts

Adversarial prompting refers to deliberately crafted inputs designed to override safeguards, extract restricted information or produce inappropriate outputs.

On the output side:

  • Responses may be inaccurate, incomplete or misleading

  • Fabricated information may be presented confidently

  • Outputs may reflect bias or inappropriate content

  • Advice may be interpreted as authoritative when it should not be

If AI-generated responses are used in tenant communications, advice, safeguarding interactions or policy interpretation, the impact can extend beyond technical error into legal and reputational exposure.

Boards should expect assurance that:

  • Clear policies exist for acceptable AI use

  • Sensitive data is not exposed through prompts

  • Higher-risk chatbot use cases are formally approved

  • Guardrails and content controls are configured appropriately

  • Human oversight exists for material interactions

  • Outputs are monitored and periodically reviewed

  • Staff are trained on limitations and appropriate use

Malicious Attacks on Models

AI systems may be targeted through data poisoning, adversarial inputs or prompt manipulation.

These risks sit at the intersection of cyber security and data integrity.

Model security and training data integrity should form part of broader cyber assurance.

Opaque Decision Influence

Even where humans remain involved, AI may shape recommendations or prioritisation in ways that are not easily explained.

Where decisions materially affect tenants or colleagues, explainability and accountability are governance requirements, not optional enhancements.


4. AI-Assisted Development and Software Governance

AI-assisted coding is now widely accessible.

Developers and non-technical teams alike can generate scripts, automations, integrations and application components rapidly. This can improve productivity and reduce development time.

It also introduces governance risk.

AI-generated code may:

  • Contain security vulnerabilities

  • Introduce insecure dependencies

  • Circumvent secure development standards

  • Be deployed outside formal change control

  • Lack clear ownership or support arrangements

The issue is not that AI writes poor code. It is that it lowers the barrier to creating and deploying software without proportionate review.

Boards should not be reviewing development practices directly. They should be confident that:

  • AI-assisted development is covered by existing secure development lifecycle controls

  • Code deployed into production undergoes appropriate review and security testing

  • Internally developed tools are visible, documented and supported

  • Change management standards apply regardless of how code was generated

  • Software in use is acquired, supported and governed in line with organisational standards

Innovation should be encouraged but development standards and governance should not be diluted.

AI-assisted code must meet the same security, assurance and accountability expectations as traditionally developed software.


5. Detection, Response and Financial Resilience

As attack speed increases, response maturity becomes more important.

Boards should have assurance that:

  • Security monitoring is active and capable of identifying confirmed incidents

  • Incidents are formally classified and escalated

  • Incident response roles and leadership are defined

  • 24 hour detection and response capability is available, internally or through a managed provider

  • Incident response plans are periodically exercised

  • Cyber insurance cover is appropriate to realistic exposure

 

Insurance is part of resilience planning, it does not replace effective controls.


6. AI in the Supply Chain

Many organisations encounter AI through supplier upgrades rather than internal development.

AI capability may be introduced through SaaS platforms, analytics providers or contractor tooling.

This can change how data is processed, how decisions are influenced and how systems interconnect.

 

Boards should expect visibility of:

  • Which critical suppliers are introducing AI functionality

  • Whether AI features influence material operational decisions

  • Whether organisational data is used for model training

  • Whether supplier assurance reviews include AI governance considerations


7. What Proportionate Board Assurance Looks Like

Boards do not need technical model metrics. They should expect structured assurance across six areas.

Identity and Access Resilience

  • Multi-factor authentication fully deployed

  • Privileged access tightly controlled

  • Regular review of high-risk accounts

Platform and Vulnerability Integrity

  • Critical vulnerabilities resolved within agreed timeframes

  • No unsupported systems in use

  • External exposure monitored and managed

Detection and Response Capability

  • Continuous monitoring capability

  • Clear incident classification and escalation

  • Defined response leadership

  • 24 hour response coverage

  • Periodic incident exercises

AI Use Visibility and Oversight

  • Documented inventory of AI use cases

  • Proportionate risk classification

  • Impact assessments where required

  • Ongoing monitoring of higher-impact systems

  • Clear accountability for deployment

Supplier and Dependency Governance

  • Formal register of critical digital suppliers

  • Periodic assurance over supplier cyber maturity

  • Visibility of AI integration within supplier services

  • Clear recovery and notification expectations

Financial and Reputational Preparedness

  • Appropriate cyber insurance limits

  • Clear regulatory escalation pathways

  • Defined communications ownership


8. Alignment with Recognised Governance Principles

International guidance such as the OECD AI Principles and the NIST AI Risk Management Framework emphasises human-centric technology use, fairness, accountability, transparency, robustness and ongoing risk management.

Board oversight should ensure that AI adoption aligns with these principles in practice, particularly where AI influences tenant-facing services or sensitive data processing.

This does not require formal adoption of a specific framework. It requires structured governance aligned to recognised good practice.


9. Legal and Regulatory Considerations

The regulatory landscape for AI is evolving and remains fragmented across jurisdictions.

 

In the UK, the government has adopted a pro-innovation and pro-growth stance. Rather than introducing a single overarching AI Act, the current approach relies on existing regulators applying established legal frameworks to AI use within their sectors.

This means AI risk is expected to be managed within:

  • Data protection law

  • Equality and discrimination law

  • Consumer and contract law

  • Sector regulatory standards

  • Corporate governance and accountability frameworks

The approach is adaptive rather than prescriptive.

 

In contrast, the European Union has taken a more precautionary path through the EU AI Act, introducing a formal risk classification regime and specific prohibited uses and obligations for high-risk AI systems. Singapore has similarly focused on governance frameworks that emphasise explainability, accountability and ethical use.

 

For housing providers in England, the Regulator of Social Housing has not identified AI as a standalone risk category. However, Boards should not interpret that as absence of oversight.

The regulator will reasonably expect that AI adoption is governed within existing expectations, particularly in areas such as:

  • Data quality and integrity

  • Accurate regulatory returns

  • Fairness in decision-making

  • Tenant safety and service standards

  • Value for money

  • Robust governance and risk management

If AI influences rent setting, repairs prioritisation, tenant communications, safeguarding flags or performance reporting, governance expectations already in place will apply.

 

The question is not whether AI is regulated separately. It is whether AI use remains aligned with existing legal and regulatory duties.

Boards should therefore expect assurance that:

  • AI use is documented and accountable

  • Data protection impact assessments are completed where required

  • Equality and fairness considerations are assessed where AI influences decisions

  • Regulatory reporting processes are not compromised by automated data flows

  • Legal and contractual terms with suppliers reflect AI processing realities

AI governance should not sit outside existing risk and assurance structures. It should be integrated into them.


Closing thoughts - taking a balanced position

Boards do not need to be alarmed by AI, but concern is warranted where:

  • AI use is undocumented

  • Supplier AI adoption is not visible

  • Data lineage and integrity are unclear

  • Core cyber controls are inconsistent

  • Detection and response capability is not well evidenced

  • Dependency risk is not visible in ARAC reporting

AI is an amplifier of opportunity and risk, it also introduces distinct model and data governance considerations.

Board oversight should therefore focus on four questions:

  1. Do we have a deliberate strategy on where we want to implement AI in our business?

  2. Do we have visibility of where and how AI is being used?

  3. Are our cyber and data fundamentals strong enough to support those uses?

  4. Are the ethics and usage clear when outcomes affect tenants, colleagues or reputation?

If those answers are clear, AI becomes an opportunity managed within guardrails, not a risk operating beyond them.

Adam Collin

Adam Collin is the founder of Done Better., a consultancy helping housing providers, charities, and SMEs make IT, cyber, and governance simpler, smarter, and more sustainable. A former CIO, CISO, and NED, Adam is passionate about making IT practical to enable real-world impact. Adam is also an ADHD UK ambassador and neurodiversity champion.

https://www.donebetter.co.uk
Next
Next

Guide - Cyber and Digital Risk Metrics for Boards