Most boards are dealing with the AI position they can see, and the risks attached to it. The bigger one is invisible: AI is already answering questions about your organisation using pages you forgot you published. Digital teams have been screaming about this for years, unheard. Now AI has made it a boardroom problem.
The World Economic Forum's Global Risks Report places misinformation and disinformation as a second-ranked global risk over 2 years, with adverse AI outcomes rising sharply over 10 years. Control of what you know, and oversight of what you own, is now a strategic matter.
Post-COVID publishing growth, distributed content updates, department self-build projects, and material left online after its purpose has passed have created a digital estate organisations have lost sight of, or worse, content the digital team was never aware of.
The Triad of Exposure names 3 distinct risks that follow from that position: misrepresentation and misinformation, future visibility and competitiveness, and the cyber problem. None requires AI to behave unusually. Each follows directly from the gap between what the organisation manages and what remains visible.
Misrepresented, outranked, or quietly exposed: you cannot fix what you have never mapped. This article and the full report set out how this exposure forms, why it sits outside what risk and governance currently measure, and the first practical move to bring it under control.
AI reads outdated material and presents it with confidence. The concern is not theoretical: it is the leading AI threat in the 2026 Gallagher benchmarking, cited by 57% of respondents.
85% of brand mentions in AI search come from third-party content. With owned material unclear, the wider picture is harder to influence, and harder to correct once it circulates.
Old domains, forgotten subdomains, one-off campaigns, and joint partner pages stay online without maintenance. The estate AI reads is the same estate attackers probe for vulnerabilities.
Prefer to read it? The full transcript of the recording is published word for word, 4 min 47 sec of audio.
"An organisation's AI readiness is measured by what AI reads across the footprint it has built online over years, much of it forgotten, not by what it believes it presents today."Lloyd Golley · Executive · CxO Bureau
01First exposure: misrepresentation & misinformation.
AI errors, misinformation and hallucinations are the leading perceived AI threat in Gallagher's 2026 AI Adoption and Risk Benchmarking, cited by 57% of respondents. That concern takes on a more specific shape when the mechanism is considered: AI reads the material that is available. If that material is outdated, contradictory, or superseded, AI will often present it as if it were current.
This is not a failure of AI capability. It is a consequence of the estate an organisation has accumulated. The more material there is that no longer accurately represents the organisation, the greater the surface area for inaccurate AI output.
Gallagher's 2026 AI Adoption and Risk Benchmarking surveyed organisations on their primary AI concerns. Misrepresentation through AI errors & hallucinations came out as the leading perceived threat.
02Second exposure: future visibility & competitiveness.
McKinsey reports that brand-owned sites account for 5 to 10% of the sources AI search references in many categories. AirOps reports that 85% of brand mentions in AI search come from third-party content. The organisation's own estate is no longer the whole picture, but unclear owned material makes the wider picture harder to influence.
Adobe reported that traffic from AI sources to US retail sites grew 393% year on year in Q1 2026. AI is not only a research layer. It is becoming part of how demand is directed, which makes accurate representation and inclusion progressively more valuable. The commercial case for governance was already forming; that figure accelerates it.
The compounding factor is time. AI capability is moving from answering to acting. As agentic AI becomes more embedded in research, procurement, and recommendation flows, the estate that is unclear today becomes harder to manage once it has already been read, summarized, and used elsewhere. Correcting a record after it has circulated costs more than managing it before it does.
03Third exposure: the cyber problem.
Unknown, unmanaged assets are an open door. Old domains, forgotten subdomains, and legacy applications that remain online are almost always unmaintained, and unmaintained software is unpatched software. The surface area that AI can read from the outside is the same surface area a motivated attacker can probe.
In June 2026, a US export control directive suspended access to Anthropic's Fable 5 and Mythos 5 models over their cybersecurity capability. The capability that drew that regulatory response is the capability now reading the estate an organisation has lost sight of. The cyber dimension is not a future concern. It is present in the same unmanaged estate that creates the first 2 exposures.
Digital teams have warned about this for years: clearer ownership, fewer platforms, stronger control, a single manageable digital estate. Post-COVID publishing growth made that harder to achieve, with more short-life content, more distributed updates, and more material left online after its purpose has passed. The backlog grew faster than the governance.
04Insurance impact: exposure, mitigation and evidence.
Insurance is not separate from the Triad of Exposure. It is one of the places where exposure becomes financially visible. Misrepresentation, misinformation and cyber exposure can each lead to questions about loss, liability, policy wording, exclusions, limits, controls, and the evidence an organisation can produce to show mitigation.
Gallagher's 2026 AI Adoption and Risk Benchmarking reports that AI errors, misinformation and hallucinations are the leading threat from AI adoption, cited by 57% of respondents. The same report found 1 in 5 insurance industry respondents had a client experience a loss or claim from AI-related risks in the past year, with just over half fully covered. Gallagher also notes that AI-related exposures are driving exclusions, endorsements, bolt-on covers, and bespoke AI policies.
| Exposure | Insurance relevance | Evidence leadership should be able to show |
|---|---|---|
| Misrepresentation | AI may present the organisation in a way leadership would not approve, creating customer, adviser, regulatory, or reputational questions. | A mapped public estate, current authoritative content, removed or corrected outdated material, and a record of decisions. |
| Misinformation | AI may use old or conflicting material to produce inaccurate answers that create loss, complaints, or dispute over what the organisation represented. | Evidence that high-risk content has been identified, prioritized, corrected, consolidated, or withdrawn. |
| Cyber exposure | Unknown assets may sit outside normal security control, which can affect risk quality and underwriting confidence. | A view of externally visible assets, ownership, retirement decisions, and reduction of unmanaged surfaces. |
Insurance is built on risk transfer after mitigation, not as a replacement for it. RUSI describes cyber insurance as a mechanism for transferring residual risk after other risk management practices have been applied. OECD analysis notes that insurers increasingly use data, analytical tools, and engagement platforms to support risk assessment and policyholder risk reduction. Evidence of action therefore matters.
Marsh frames generative AI as a risk and insurance challenge because it can amplify misinformation, technological error and hallucinations at scale. Its guidance says AI can create and distribute misinformation faster than humans, while chatbot hallucinations can provide incorrect customer information about products and services. Marsh links this to operational, legal, regulatory and insurance exposure.
Alliant frames AI misinformation as a real-world liability issue, not a theoretical technology concern. Its AI in Insurance commentary cites fabricated AI-generated legal cases reaching court, then states that AI information cannot be fully trusted without human vetting. Its governance guidance links responsible AI use to accountability, transparency, security, compliance and reputational protection.
Organizations that can demonstrate they are identifying, assessing, and reducing exposure are better placed in underwriting and renewal discussions than organisations relying on assertion. No outcome is automatic. The value lies in having evidence of mitigation, which gives executives a stronger basis to discuss risk quality, coverage terms, limits, and exclusions.
05The scale that matters: 41%.
P&C/Sitemorse risk profiling, covering more than 119 million websites across the period 2017 to 2023, indicates that 41% of an organisation's digital footprint is unknown to its own digital teams. That is not a web management statistic. It is a measure of the exposure AI can read from the outside that the organisation itself cannot see.
Of the total digital footprint, 41% is unknown to digital teams.
The remaining 59% sits within the managed estate.
That 41% is the part of the estate where misrepresentation sits unchecked, where old content shapes AI outputs that no one inside the organisation has authorized, and where the cyber exposure is most acute. It is also, practically, where a governance program begins: not with a major 8-month consultative exercise, but with mapping what is actually there.
06Why this belongs at executive level.
Misrepresentation can affect trust. Misinformation creates regulatory issues. Weak visibility affects future demand. Maintaining material that no longer supports the organisation can keep cost in the system while increasing the work needed to correct it, and that cost compounds the longer it is left.
The 3 exposures do not sit in a single department. Misrepresentation touches communications and legal. Visibility affects commercial teams. The cyber dimension involves IT, risk, and the board. A governance response that sits below executive level will not reach across all 3. The scope of the problem matches the scope of who needs to own the answer.
This is also a timing issue. The AI decision should not be framed as what to build next. It should start with what exposure the organisation is already carrying. The estate that is unclear today is being read now, by AI systems that will carry what they find into outputs, recommendations, and decisions. The ground does not pause while the program is planned.
07World Economic Forum: global impacts of AI.
The World Economic Forum identifies false or misleading information as a material risk, with AI increasing both the volume of content and the difficulty of distinguishing accurate information from inaccurate material. Misinformation can result from AI-hallucinated content, commonly caused by foundation issues such as broken links to key references, or from human error, making the issue relevant beyond deliberate campaigns.
For organisations, this creates a governance exposure wherever information is published, reused or trusted without current oversight. Out-of-date pages, unmanaged content, weak ownership and poor review controls can allow inaccurate information to persist, then be repeated by users, search engines, AI systems and external advisers.
WEF's cybersecurity reporting adds a connected exposure: expanding supply chains reduce visibility, increase attack surfaces and make third-party vulnerabilities harder to control. That matters because compromised digital assets, software dependencies or supplier systems can become trusted routes for hostile or inaccurate content. The practical lesson is clear: organisations need current visibility over their digital estate.
08Immediate steps, not 8 months of planning.
The first step is not a major program. AAAnow offers a Discovery and Mapping answer. 2 days of client resource to start; the work continues remotely. Within 90 days, leadership can see what exists, where exposure sits, and what should be kept, corrected, or removed.
AAAnow's relevance is that it starts where the exposure starts: outside-in, with the estate AI can already read. The value is evidence of what is visible, what is unknown, and where the organisation is carrying unnecessary risk, mapped against 25 years of digital assessment and 3.7 trillion data points.
The AI readiness decision begins with what is already there, not what is planned next.
Sources.
- McKinsey & Company. New front door to the internet: Winning in the age of AI search. Supports the 5 to 10% brand-owned AI search source reference point. mckinsey.com/capabilities/growth-marketing-and-sales/our-insights/new-front-door-to-the-internet-winning-in-the-age-of-ai-search (opens in a new tab)
- AirOps. The Influence of Offsite Signals in AI Search. Supports the 85% third-party brand mention point. airops.com/report/the-influence-of-offsite-signals-in-ai-search (opens in a new tab)
- arXiv. UDA: A Benchmark Suite for Retrieval Augmented Generation in Real-world Document Analysis. arxiv.org/abs/2406.15187 (opens in a new tab)
- arXiv. Revolutionizing Retrieval-Augmented Generation with Enhanced PDF Structure Recognition. arxiv.org/abs/2401.12599 (opens in a new tab)
- arXiv. Capturing Logical Structure of Visually Structured Documents with Multimodal Transition Parser. arxiv.org/abs/2105.00150 (opens in a new tab)
- arXiv. Understanding the Logical and Semantic Structure of Large Documents. arxiv.org/abs/1709.00770 (opens in a new tab)
- arXiv. Structured Linked Data as a Memory Layer for Agent-Orchestrated Retrieval. arxiv.org/abs/2603.10700 (opens in a new tab)
- Section508.gov. Create accessible PDFs. section508.gov/create/pdfs/ (opens in a new tab)
- Government Digital Service. Why GOV.UK content should be published in HTML and not PDF. gds.blog.gov.uk/2018/07/16/why-gov-uk-content-should-be-published-in-html-and-not-pdf/ (opens in a new tab)
- CACI. Website sprawl: the cost of CMS fragmentation. caci.co.uk/blog/website-sprawl-cost-cms-fragmentation/ (opens in a new tab)
- Gallagher. The 2026 AI Adoption and Risk Benchmarking survey. Supports the 57% concern figure for AI errors, misinformation and hallucinations as the leading AI threat, and the 1-in-5 insurance industry finding. ajg.com/news-and-insights/features/ai-adoption-and-risk-benchmarking-2026/ (opens in a new tab)
- OECD. Leveraging technology in insurance to enhance risk assessment and policyholder risk reduction. Supports the point on insurers using data and analytical tools to support risk assessment. oecd.org/finance/leveraging-technology-in-insurance-to-enhance-risk-assessment-and-policyholder-risk-reduction.htm (opens in a new tab)
- RUSI. Cyber Insurance and the Cyber Security Challenge. Supports the point that cyber insurance transfers residual risk after other risk management practices have been applied. static.rusi.org/247-op-cyber-insurance-fwv.pdf (PDF, opens in a new tab)
- Marsh. Debunking Generative AI myth #3: GenAI insurance issues. Generative AI can amplify misinformation and hallucinations. marsh.com/en/services/cyber-risk/insights/gen-ai-three-myths-ai-insurance.html (opens in a new tab)
- Marsh. Two years after ChatGPT: The evolving world of generative AI, risk, and insurance. Chatbot hallucinations can provide incorrect customer information about products and services. marsh.com/en/services/cyber-risk/insights/generative-ai-evolving-considerations.html (opens in a new tab)
- Alliant. AI in Insurance: Revolutionizing the Industry. Unchecked AI information can create real-world liability. engage.alliant.com/trendsandviewsQ323/article5-503T1-214633.html (opens in a new tab)
- Alliant. Establishing a Governance Framework for AI Risk Management. AI governance must address accountability, transparency, security, compliance and reputational damage. alliant.com/news-resources/article-establishing-a-governance-framework-for-ai-risk-management/ (opens in a new tab)
- World Economic Forum. Global Risks 2025: a world of growing divisions. Supports the points that false or misleading content is rising, AI makes it harder to distinguish accurate from inaccurate information, and misinformation can result from AI-hallucinated content or human error. weforum.org/publications/global-risks-report-2025/global-risks-2025-a-world-of-growing-divisions-c943fe3ba0/ (opens in a new tab)
- World Economic Forum. 5 risk factors from supply chain interdependencies. Supports the points on limited supply-chain visibility, growing attack surfaces, system interdependencies, third-party vulnerabilities, and weaker control over supplier security maturity. weforum.org/stories/2025/01/5-risk-factors-supply-chain-interdependencies-cybersecurity/ (opens in a new tab)
- World Economic Forum. Global Cybersecurity Outlook 2025: understanding complexity in cyberspace. Supports the points on third-party software vulnerabilities, lack of ecosystem visibility, uncertainty across dependencies, AI-related vulnerabilities and the need to understand organisation-specific cyber risks. weforum.org/publications/global-cybersecurity-outlook-2025/in-full/1-understanding-complexity-in-cyberspace-587e8c5eba/ (opens in a new tab)