Businesses Want Insurance for Generative AI Risks - What Carriers Should Build Next
Published: October 14, 2025
Generative AI is moving from pilot to production across core business functions. That brings uplift in efficiency and new exposure categories that don't fit cleanly inside today's cyber or E&O products.
A new Geneva Association report, based on 600 corporate insurance decision-makers across China, France, Germany, Japan, the UK, and the US, shows strong buyer intent and clear pain points. For insurers, this is an opening to lead with coverage architecture, risk engineering, and partnerships.
What buyers are signaling
- Adoption: 71% of surveyed businesses have implemented GenAI in at least one function.
- Demand: Over 90% want insurance protection for GenAI risks; two-thirds would pay at least 10% higher premiums.
- Top concerns: Cybersecurity first, then third-party liabilities (IP/copyright, defamation, misinformation) and operational disruption.
- Market friction: Verification of controls and tail risk size feel similar to early cyber markets.
- Carrier response: Testing policy extensions and standalone AI cover; modular structures and cross-sector partnerships rank as the best path to close protection gaps.
Implications for carriers and brokers
- Move fast on modular offerings: combine named-peril grants for AI output errors, training-data IP perils, prompt-injection/jailbreak incidents, and AI-driven business interruption.
- Avoid silent AI: clarify intent across cyber, tech E&O, media liability, product liability, and D&O. Add carve-backs where appropriate.
- Underwrite the AI stack, not just the enterprise: model providers, fine-tuning vendors, data pipelines, prompts, guardrails, and human-in-the-loop checkpoints.
- Price for correlated events: prepare for model-wide defects, widespread jailbreak waves, or supplier outages that hit many insureds at once.
- Pair coverage with services: pre-bind assessments, red-teaming, copyright risk scans, and incident playbooks to improve outcomes and selection.
Coverage blueprint to test
- Third-party liability: IP/copyright infringement from training or outputs; defamation; privacy violations; misleading AI outputs causing client loss.
- First-party: incident response, forensic costs, data rectification, model recall/suspension costs, revenue loss from AI-driven outages or unsafe-output shutdowns (with waiting periods).
- Security events: prompt injection, model poisoning, data leakage via prompts/plugins, supply-chain breaches across AI tooling.
- Wording anchors: define "AI System," "AI Output," "Authorized Use," and "Adverse AI Event." Tie triggers to verifiable events and specified uses.
- Levers: sublimits per module, aggregates, coinsurance, waiting periods, retro dates for training-data exposures.
Underwriting checklist (practical)
- Use cases: customer-facing vs internal; decision automation vs assistive; safety criticality.
- Models: which base models; update cadence; fine-tuning methods; RAG vs pure generation.
- Controls: content filters, classification, watermarking, toxicity/PII screens, HIL sign-off for high-impact actions.
- Data governance: provenance checks, licensing posture, opt-outs, dataset audit trails, retention and deletion.
- Security: prompt-injection defenses, plugin/API allow-lists, secrets handling, model access isolation, vendor risk tiers.
- Monitoring: output logging, feedback loops, red-team results, incident thresholds and rollback plans.
- Legal: disclaimers and user notices, indemnities with AI vendors, IP clearances, record of consent.
Pricing and accumulation
- Use scenarios over pure historical loss: simulate model flaw, provider outage, jailbreak campaigns, and dataset copyright waves.
- Segment by controls and use case criticality; price credits for strong governance and verified guardrails.
- Manage tail: sublimits for IP and model recall; aggregates across shared suppliers; reinsurance with AI-specific event definitions.
Claims and wording guardrails
- Trigger clarity: what qualifies as "defective output," how causation is evidenced, and which environments are covered (prod, pilot, shadow).
- Exclusions to consider: intentional misuse, illegal data acquisition, bodily injury/property damage unless endorsed, and war/critical infrastructure events.
- Carve-backs: for insured's reasonable reliance on AI with documented controls; for good-faith use of licensed datasets or vendor-backed indemnities.
Distribution and partnerships
- Bundle coverage with assessments from AI security firms and IP counsel; certify controls at bind and renewal.
- Partner with model and platform providers for telemetry and incident data to reduce verification costs.
- Broker playbook: map client AI usage, align modules to exposures, and stage limits with control milestones.
Signals from industry leaders
"Few technologies in history have spread as fast as Gen AI, yet its risks are complex and poorly understood," said Jad Ariss, Managing Director of the Geneva Association. "This report provides insurers with a clearer picture of business demand for insurance protection, helping the industry anticipate needs. Insurers have a unique role to play in ensuring that Gen AI adoption is safe and sustainable."
Ruo (Alex) Jia, Director of Digital Technologies at the Geneva Association and lead author of the report, noted that GenAI "amplifies some existing risks and creates entirely new categories of exposure that extend beyond traditional insurance boundaries." He added that insurers must define clear risk parameters and test modular coverage models to meet evolving needs.
About the Geneva Association
The Geneva Association is a global association of insurance companies, bringing together CEOs from leading insurers and reinsurers. Members are headquartered in 26 countries, manage USD 21 trillion in assets, employ over 2.5 million people, and provide protection to 2.6 billion individuals.
Further reading
Upskill your insurance team on GenAI
If you're building AI cover or advising clients, structured training speeds up product design and control verification. Explore role-based programs here: Complete AI Training - Courses by Job.
Your membership also unlocks:
