How GenAI risks are changing E&O insurance
25 November, 2025 | 3 minute read
GenAI now touches multiple lines, and that's forcing hard choices across the Errors & Omissions market. The core challenge: carriers need to address AI risk without taking on open-ended exposure.
Some have gone blunt. "Absolute AI Exclusions" are being added to policies, cutting out any claim involving the use, development, or deployment of AI. Berkley, for example, has applied this approach on D&O, E&O, and Fiduciary Liability. Others are moving in the opposite direction-offering targeted coverage based on how well an insured governs AI.
Why classic E&O struggles with GenAI
Traditional E&O was built for human mistakes. GenAI introduces algorithmic failure, scale, and speed-categories that spill beyond a single professional and into systems risk.
A recent Lloyd's Market Association report highlights the fault lines and where disputes may arise, especially around causation and aggregation. You can find the LMA here: Lloyd's Market Association.
Key risk themes to watch
- Hallucinations: Professionals relying on GenAI outputs risk giving incorrect advice. If that reliance is unreasonable or undocumented, liability follows.
- Confidentiality and data handling: Feeding client information into public models can breach confidentiality and data protection duties, triggering regulatory action and civil claims.
- Systemic risk and aggregation: Repeated errors from a shared AI tool look less like a one-off professional misstep and more like a software issue. That pushes losses toward IT/cyber coverage and creates tough calls under aggregation and series clauses.
How products are shifting
The market response splits into three tracks. First, broad exclusions remove AI entirely. Second, endorsements bring AI back in under tightly defined conditions (e.g., documented controls, human review, approved vendors). Third, standalone AI liability products aim to price and ring-fence these exposures.
The goal is clarity: define responsibility between insured and insurer, and align coverage with how AI is actually used and controlled inside the firm.
Pricing and wording: where this is heading next
Expect clearer policy language around AI usage, data flows, and systemic triggers. Pricing will lean more on governance quality than on sector averages.
Carriers will model platform-level failures that could spark many claims at once and refine aggregation wording to keep exposure within intended limits.
What insured firms can do now to secure better terms
Underwriters are moving past "Do you use AI?" to "How do you control it?" Strong governance will lower friction on placement, pricing, and coverage scope.
- Map where GenAI is used across the business, who owns it, and the purpose for each use case.
- Adopt clear acceptable use policies for staff and third parties; log approvals for new AI tools.
- Train teams on prompt discipline, data sensitivity, and model limitations; refresh training quarterly. For structured upskilling, see AI courses for teams.
- Keep sensitive data out of public models; use approved environments with access controls and audit trails.
- Implement human-in-the-loop reviews for client-facing outputs; record who reviewed and what changed.
- Vet vendors: security posture, indemnities, service levels, and incident notification timelines.
- Monitor and test models for drift, bias, and error rates; set thresholds that trigger escalation.
- Run incident response playbooks for AI-related errors, data leaks, and takedown requests.
- Document everything-controls, workflows, approvals, and exceptions. This is what underwriters will want to see.
- Coordinate coverage across E&O, cyber, tech E&O, and media liability to reduce gaps and disputes.
What carriers and brokers should lock in
Distribution and claims teams need a consistent view of AI risk to avoid surprises at FNOL or during aggregation analysis.
- Standardize AI underwriting questionnaires by segment and exposure type (professional advice, software, data, media).
- Link rating factors to control maturity: data governance, model oversight, vendor management, and auditability.
- Offer a menu of endorsements that restore cover under defined conditions; state triggers and carve-backs plainly.
- Clarify boundaries across lines (E&O vs. cyber/IT) to reduce gray areas on systemic incidents.
- Strengthen aggregation and series wording for AI-driven repeat errors and platform outages.
- Align claims protocols for AI incidents, including evidence standards for human review and vendor fault.
- Update reinsurance terms and event definitions to reflect model-wide or platform-level failures.
Bottom line
GenAI pressure-tests classic E&O assumptions. Firms that can prove disciplined governance will win on price, clarity, and resilience.
For everyone else, exclusions and disputes will fill the gap. The fix is straightforward: show your controls, document your process, and make it easy for the underwriter to say yes.
Your membership also unlocks:
