B2B AI PlatformB2B AI Platform
AI Security: The New Boardroom Priority
B2B AI Platform

AI Security: The New Boardroom Priority

Emma RobertsWords by Emma Roberts

The measured size of it

58%
Respondents citing cyber threats as top concern
70%
Cyber insurance buying

Every board that has accepted AI security as its own problem now faces a split decision, and the two paths are paid for from different pockets.

Path one is perimeter defence. An attacker uses AI to map your estate faster, draft a phishing message in your finance director's register and find the host that missed a patch. You answer with controls you can buy: identity, segmentation, filtering, logging, response. Vendors sell this. It has a maturity model, a budget line and a certificate for the wall.

Path two is internal conduct. Staff paste a contract into a chatbot. A supplier's assistant ingests your price list. A model bought last quarter answers customers with no record of what it was told or who approved it. You answer with authority, inventory and evidence. No single vendor sells the whole. There is no certificate. There is a named owner, and that owner sits in the boardroom.

The 2026 Travelers Risk Index, reported by Risk & Insurance, shows how far each path has been walked. The first is being funded. The second is being discussed.

Cyber tops the survey again, and the ranking is not the finding

Cyber threats led the Travelers list again, cited by 58% of respondents, ahead of economic uncertainty and medical cost inflation, Risk & Insurance reported. It is the fifth time in eight years that cyber has taken the top spot.

That is a mature answer to a problem that keeps changing shape. A concern that wins most years stops being read closely, and the useful material sits underneath it. An AI-related security breach or cyber event ranked second among specific cyber concerns at 55%, and hackers gaining unauthorised access ranked first at 56%, the same report found. Read the two numbers together. Business leaders no longer file AI under emerging technology and hacking under security. They file them side by side.

Half the AI worry list describes your own conduct

The top AI-specific concern is external: attackers using AI to find and exploit vulnerabilities, at 56%, followed by AI-generated phishing, deepfakes and social engineering at 54%, according to Risk & Insurance.

Then the list turns inward. Sensitive company data retained to train external models was cited by 52%, employees putting sensitive data into unsanctioned tools by 51%, third-party vendors misusing AI by 50% and inaccurate AI output driving poor decisions by 48%, the report states. Income loss from a supplier cyber incident and unauthorised access to financial accounts were each cited by 53%, it adds.

Roughly half of what worries boards about AI is not an attacker at all. It is the organisation's own use, its suppliers' use and the reliability of the output.

Why an authority gap exists between AI use and formal governance

Day-to-day AI use in the workforce stands at 89% while formal practices governing that use reach 59%, a 30-point gap, Risk & Insurance reported. That is not a training problem. It is an authority problem. Nobody has been given the job of saying which tools are allowed, what data may enter them and who signs off when a model touches a customer or a payment.

John Menefee, Travelers' vice president and enterprise cyber lead, told Risk & Insurance that the gap "really stands out", and said companies should be as focused on the risks from their own AI use as on what criminals do with the technology.

How cyber concerns vary by sector and why generic board packs fail

Travelers found that banking, construction and technology companies put unauthorised access to financial accounts first, while healthcare and manufacturing pointed to security breaches, Risk & Insurance reported. Nonprofits were most concerned about employees putting information or systems at risk. Professional services and retail named an AI-related security breach as their leading cyber concern.

A board pack written for the average company describes nobody's company. The sector breakdown is the part of the survey to read first, because it tells you which of the two paths your peers are already paying for.

Why purchasable security controls fail to close the governance gap

Cyber insurance buying reached 70%, up seven percentage points in a year, Risk & Insurance reported. Multifactor authentication for administrative users rose to 72%, onboarding and offboarding protocols to 74%, employee background checks to 79%, and firewalls and backups to 81% according to a Travelers Survey.

Every one of those is a line item. You can buy it, install it and count it. None of it tells you which team is running a summarisation tool over client data.

The same pattern runs through the boardroom. Diligent's survey of US public company directors found 50% use generative AI for meeting preparation, 39% for intelligence summarisation and 26% for benchmarking, while 13% use it for predictive analysis and 13% for real-time risk monitoring, Diligent reported. Only 2% said they had ruled AI in the Boardroom out for board work.

A board that uses AI to prepare its meetings is not thereby equipped to assure real-time risk monitoring across the enterprise. The tooling arrived ahead of the oversight.

Governance has to be built as a sequence, not a policy

A policy document proves intent. It does not prove control. The board needs five things, in this order.

  1. An inventory. Every AI use, including the ones inside software you bought and never called AI. The NCSC's guidance for leaders is clear that managers do not need to be technical, but they do need to know where the systems are and what they touch.
  2. A named owner for each material use. One person accountable for the outcome, even when the model comes from a third party. Alpha's board brief on AI security risk puts the principle plainly: making AI everyone's responsibility makes it no one's (boardbrief).
  3. Decision rights. What the system may decide alone, what needs human approval, who can suspend it and what the record shows afterwards. Ashurst's 2026 board priorities note that agentic tools, which initiate actions rather than suggest them, put those arrangements under stress (Ashurst).
  4. Contract terms with suppliers. Whether your data trains their model, how incidents are disclosed, what happens on exit. ITBrief reported that SailPoint found most organisations using AI agents lack identity tools to govern them, which is a supplier question as much as a technical one (ITBrief).
  5. An exercise, then evidence. Run one scenario: an AI-generated impersonation of the finance director authorises a payment. Time the detection and the recovery. Report what happened, not what the policy says.

Standards reach the decision or they do not reach the board

Frameworks are multiplying. Invenia's review of 2026 standards sets ISO 42001 against the NIST AI Risk Management Framework and lays out the NIST functions as govern, map, measure and manage (Invenia). The SANS Institute's draft guidelines argue for layered defences and governance that adapts as models change (SANS).

None of this substitutes for the board deciding where it will accept risk. Ashurst's read of 2026 is that the question has moved from whether frameworks exist to whether they are implemented and defensible in practice (Ashurst).

That is the test. Not a certificate or a policy signed at the away day, but a clear answer to three questions: which AI decisions could hurt us, who owns each one, and how would we know it had gone wrong?

Why the CISO is moving into the boardroom

Perimeter spending is not wasted. Attackers are using AI well, and the survey shows boards know it. But the same survey shows the internal path is underfunded relative to its risk: near-universal use, a governing practice that reaches just over half, and a worry list where half the entries describe your own people and suppliers.

The security leader is already being pulled into the room. Mexico Business News reported that AI agents are pushing CISOs into boardroom decision-making, and that Gartner data cited by CNBC projects a 6% rise in security budgets for 2026 (Mexico Business News).

Use that money on both paths, then measure the second one honestly. Count the AI uses you know about. Count the ones with a named owner. Count the supplier clauses you have signed. If those three counts are not moving, the board has bought the perimeter and left the inside open.

Sources

Source: Cyber Threats Top Business Concerns Again as AI Reshapes Risk, Travelers Survey Shows, Risk & Insurance

Changed

Recently filed

More on this

AI agents: The new sales engine
AI agents: The new sales engine
Can you actually control your AI?
Can you actually control your AI?
Can Small Firms Outpace Giants With AI?
Can Small Firms Outpace Giants With AI?