AI cybersecurity for business leaders starts with three decisions: which risks you need to reduce, which actions AI may take, and what evidence will show that protection has improved.

A faster investigation can be useful. A confident recommendation can also be wrong. Before investing, leaders need to understand both the work a tool can support and the checks surrounding it.

My perspective comes from connecting the projects recognised through the AI Awards with conversations on the AI Ireland Podcast. The projects show where organisations are applying AI. The interviews let me explore practitioners’ thinking. I then translate those examples into questions a leadership team can use.

The 2026 cybersecurity finalists and two recent podcast conversations point towards a practical starting point: ask to see the evidence behind a security decision.

What does AI cybersecurity mean for a business?

It helps to separate three conversations that can become mixed together:

Area Business question
AI used by attackers Are our protections keeping pace as familiar attacks become easier to carry out?
AI used in defence Can it help our team investigate, assess or respond more effectively?
Security of our own AI Could an AI tool expose information, follow misleading instructions or take an unwanted action?

The UK’s National Cyber Security Centre warns that AI can increase the speed and scale at which existing weaknesses are exploited. Its guidance for leaders and board members also stresses security fundamentals and understanding the access given to AI tools.

My recommendation is to identify which of these three problems a proposed investment addresses. Buying an AI-enabled security product does not, by itself, answer questions about every AI tool your employees use.

Four AI Awards examples—and the decisions they raise

BH Consulting, Disseqt AI, the Health Service Executive and Island Networks are finalists for Best Application of AI in Cybersecurity. The project descriptions below come from the published AI Awards 2026 shortlist. They describe approaches, rather than independently verified performance results. The leadership questions are my interpretation.

1. HSE: make supplier assessments traceable

The HSE’s project uses AI to assist the review of suppliers’ cybersecurity evidence during technology procurement, retaining specialist human review and traceable justifications.

This is a useful example because security decisions also happen when an organisation buys technology.

I would ask a procurement team to distinguish three things: what the supplier claims, what the documents support and what remains unanswered. A polished summary should not turn missing evidence into an approval.

Leadership question: Can we trace each important conclusion to a current source and a named reviewer?

2. Island Networks: connect investigation with business context

Island Networks’ multi-agent platform connects security tools and information sources to support investigations, correlate threats and explain recommendations, with people retaining operational oversight.

The lesson I take is to examine the route from a signal to a decision. What happened? Which service could be affected? What does the team recommend doing?

For example, disabling an account may protect a system while interrupting an important business activity. The decision needs enough context for the person responsible to weigh those consequences.

Leadership question: Does the recommendation explain the evidence, uncertainty and likely operational impact?

3. Disseqt AI: examine the AI workflow itself

Disseqt AI’s finalist project concerns security and governance for AI-powered financial workflows. It checks outputs and supports oversight through monitoring, explainable recommendations and auditable records.

This brings the third conversation into focus: the security and behaviour of the AI system your organisation is introducing.

A demonstration using expected inputs is only one part of an assessment. I would also want the team to explain how they test inaccurate information, unusual requests and attempted misuse—and what happens when a check fails.

Leadership question: What evidence shows that our AI workflow behaves acceptably when conditions are difficult?

4. BH Consulting: make ownership clear

BH Consulting’s BH Haven combines cybersecurity, data protection and AI governance, using AI-supported automation alongside expert oversight to help smaller organisations manage digital risk.

For a smaller business, my question is who owns the work across those areas. A service can bring expertise together, but the customer still needs to understand responsibilities, escalation routes and unresolved issues.

Leadership question: If a finding falls between security, privacy and AI use, who makes sure it is addressed?

What the podcast conversations add

Raluca Saceanu: keep the fundamentals in view

In episode 251 with Smarttech247 CEO Raluca Saceanu, we discussed AI’s role in attack and defence, knowledge of your own environment and security by design. The published episode summary captures her point:

“No matter how advanced cybersecurity or AI gets, the fundamentals still apply.”

My takeaway for leaders is to test whether a new investment strengthens the organisation’s existing security work. It should be clear how findings reach the people who can act on them.

Cyril Treacy: ask for evidence as well as policy

In episode 252 with Disseqt AI co-founder and COO Cyril Treacy, the discussion covered the distinction between governance and assurance, testing and continuing oversight as AI systems change.

I translate that into two requests: show me the rule, then show me how you checked that the system follows it.

A policy might require approval before a consequential action. The next question is whether that approval is enforced in the workflow and recorded when it happens.

Seven questions to ask before investing in AI cybersecurity

  1. Which business risk are we addressing? Name the service, information or decision you need to protect. “We need more AI” is not a defined security requirement.
  2. What will change in the team’s work? Specify whether AI will summarise, prioritise, recommend or act. Identify who receives the result and what they do next.
  3. What evidence can a reviewer inspect? Ask for source records, relevant context and a clear indication of missing information. Check whether an explanation is supported by those records.
  4. What access does the product require? Review the information it receives, where it is processed, who can access it and which connected systems it can affect.
  5. How will we test failure? Agree realistic difficult cases with your security specialists. Include false alarms, missed issues and attempts to misuse the AI workflow.
  6. How will we measure the complete result? Include review, corrections and operational disruption. Faster first-pass analysis is useful only when the overall result is acceptable.
  7. Who owns the decision after deployment? Set responsibility for monitoring, escalation, changes and stopping the system. Decide when the evidence needs to be reviewed again.

For more detail, see my guides to AI agent permissions and measuring AI value. These questions help structure a buying decision; the technical assessment belongs with the people responsible for security.

A practical test: can the system recognise missing evidence?

Consider an illustrative procurement exercise. This is a suggested test, not a reported result from an Awards finalist.

Your team gives an AI assistant a supplier document pack containing an out-of-date policy, an unanswered question and conflicting descriptions of access controls. Ask it to produce a draft assessment with supporting references.

The reviewer should check whether it identifies the gaps, points to the right documents and avoids inventing reassurance. Record errors and the time required to correct them. Agree in advance that the assistant cannot approve the supplier.

Your security team should also assess how the tool handles untrusted material. OWASP’s prompt-injection guidance explains that instructions inside external content can influence an AI system’s behaviour. It recommends measures including restricted privileges, separation of untrusted content and human approval for high-risk actions.

My buying test is straightforward: can the team show both useful results and effective limits using evidence from an appropriate trial?

Questions business leaders often ask

Can AI replace the security team?

I would assess it against particular tasks. Ask which work it can support, what expertise remains necessary and who is accountable when its output is wrong. The finalist examples here retain human expertise or oversight.

Where should a smaller business start?

Start with a discussion between the business owner and whoever manages security. Identify an important service, its current protections and the unresolved problem. Then decide whether AI is relevant to that problem and what a controlled evaluation would involve.

Bring an evidence-based AI discussion to your leadership event

The value I bring is connecting live examples with the decisions leaders need to make. The AI Awards provide project perspectives. Podcast conversations add practitioner judgement. Together, they give an audience concrete material to question and apply.

For your next leadership event, an AI keynote can help the audience understand opportunities, challenge claims and agree useful next steps with their technical teams.

Discuss an AI keynote for your leadership audience.

Share This Story, Choose Your Platform!

Planning an event or leadership offsite?

Mark Kelly delivers evidence-based AI keynotes and hands-on AI leadership workshops for boards, executive teams and conferences across Ireland and Europe.

Check Mark’s availability