Direct comparison
AI Liability Insurance: Who Underwrites the Risk
Armilla's six AI liability coverage lines, its Lloyd's underwriter roster, and the MKIII case study, mapped to CASRAI's NIKOLAI residual-risk element.
Written and maintained by CASRAI Editorial Board
Last updated
Ask CASRAI · free to try
Ask about AI Liability Insurance: Who Underwrites the Risk
Ask your first 2 questions free below. Subscribers get 150 a day for $29 a month.
Ask CASRAI answers research-administration questions and cites the passages behind every claim. When our sources don't cover a question, it says so.
Answers draw on CASRAI's guides and dictionary plus the federal and funder documents we index: Federal Register, Grants.gov, Regulations.gov and UKRI.
Works on this site and inside Claude, Cursor and the AI tools you already use.
Everything CASRAI publishes — this page, the dictionary, the guides and the news — stays free to read, with no account and no card.
How do AI Model Error Liability, AI Model Output Liability, AI Agent Mistakes, Non-Breach Privacy & Data Leakage, AI Model Property Damage, AI Model Regulatory Violations compare side by side?
The table below compares AI Model Error Liability, AI Model Output Liability, AI Agent Mistakes, Non-Breach Privacy & Data Leakage, AI Model Property Damage, AI Model Regulatory Violations across 4 procurement-relevant dimensions, from what it responds to through underwritten by.
Side-by-side comparison
| Dimension | AI Model Error Liability | AI Model Output Liability | AI Agent Mistakes | Non-Breach Privacy & Data Leakage | AI Model Property Damage | AI Model Regulatory Violations |
|---|---|---|---|---|---|---|
| What it responds to | Financial loss from the AI's own underperformance, errors, or hallucinations | A third party's legal claim over a defective AI output | A claim arising from an underperforming AI agent, including an escalation failure | Unintended disclosure of personal or confidential data through model outputs | Third-party physical property damage caused by an AI system's failure | Defense costs and fines from an AI-related regulatory action |
| Example trigger | Accuracy or output quality falls below what was represented, causing measurable financial loss. | A defamation or confidentiality-breach claim over content the model produced. | An autonomous agent fails to escalate a decision it should have routed to a human. | Memorized training data, or other personal data, surfaces in a model output — no hack required. | An AI-controlled physical system damages third-party property. | A regulator opens an enforcement action over an AI system's conduct. |
| Loss falls on | The insured's own financial loss | A third party's claim against the insured | A third party's claim against the insured | A third party's privacy claim against the insured | A third party's property claim against the insured | The insured's regulatory defense costs and fines |
| Underwritten by | Lloyd's markets via Armilla as coverholder (Chaucer, AXIS Capital, Convex, Swiss Re, Greenlight Re) — program-level roster; no public per-line breakdown | Lloyd's markets via Armilla as coverholder (Chaucer, AXIS Capital, Convex, Swiss Re, Greenlight Re) — program-level roster; no public per-line breakdown | Lloyd's markets via Armilla as coverholder (Chaucer, AXIS Capital, Convex, Swiss Re, Greenlight Re) — program-level roster; no public per-line breakdown | Lloyd's markets via Armilla as coverholder (Chaucer, AXIS Capital, Convex, Swiss Re, Greenlight Re) — program-level roster; no public per-line breakdown | Lloyd's markets via Armilla as coverholder (Chaucer, AXIS Capital, Convex, Swiss Re, Greenlight Re) — program-level roster; no public per-line breakdown | Lloyd's markets via Armilla as coverholder (Chaucer, AXIS Capital, Convex, Swiss Re, Greenlight Re) — program-level roster; no public per-line breakdown |
Common questions
Common questions about AI Model Error Liability vs AI Model Output Liability vs AI Agent Mistakes vs Non-Breach Privacy & Data Leakage vs AI Model Property Damage vs AI Model Regulatory Violations
Is Armilla the only insurer writing affirmative AI liability coverage?
+
No. Armilla is one MGA/Lloyd's coverholder in a still-forming market. CASRAI's own guide on why no US state mandates this coverage documents other carriers staking out adjacent, differently-structured pieces of the same broader AI-risk frontier — this page is scoped specifically to Armilla's named six-line product and MKIII case study, not a survey of every carrier active in the space.
Does buying this insurance satisfy California SB 53's accountable-decision-maker requirement?
+
No — they are different mechanisms answering different questions. SB 53's requirement, tied to CASRAI's own NIKOLAI Commitment element in Track N9, is about who internally reviews and signs off on deploying a frontier model. An insurance policy is a separate, external instrument that transfers financial risk after that decision has already been made; it does not substitute for, or satisfy, an internal governance sign-off requirement.
What's the difference between AI Model Error Liability and AI Model Output Liability?
+
AI Model Error Liability responds to the AI simply performing badly — an accuracy shortfall or hallucination causing financial loss, with no third-party claim required. AI Model Output Liability is triggered only once a third party has a legal claim over what a specific output said or did, such as defamation or a confidentiality breach.
Is affirmative AI liability insurance a new product launched in 2026?
+
No dated 2026 launch announcement was found for Armilla's program during CASRAI's research for this page, so this page does not make that claim. It presents the six coverage lines and the named underwriter roster as current-state fact, verified directly against Armilla's own website on September 20, 2026, not as a 2026 product launch.
Going deeper
