Armilla, a Lloyd’s coverholder that sells standalone insurance for failures of artificial intelligence, lists limits of “up to US$25 million aggregate per insured, and up to US$5 million per model” on its policy page, published on 14 October 2026. The risks it names include “AI Agent Mistakes”, defined as “claims arising from agent actions, including incorrect decisions and failures to escalate.”
The page went up in a week when brokers and insurers debated whether to cover AI at all. Marsh CEO John Doyle said a broad-based AI exclusion in insurance policies could create “real confusion and uncertainty,” Reuters reported in a story Insurance Journal published on 16 October 2026; the same story noted OpenAI and Anthropic have disclosed AI agents behaving unexpectedly.
Business Pill · THE AI AUDIT PROBLEM
A short explainer on the AI audit problem: when making work is cheaper than checking it. It teaches the general idea only and says nothing about any company in this story.
The key insight: As we read it, AI risk is getting a price tag from specialists before the wider market has decided whether to cover it: Armilla lists a US$25 million limit while the specialty executives Business Insurance spoke to are not adopting ISO’s optional generative-AI exclusions.
Armilla says its affirmative AI liability insurance “is underwritten by certain underwriters at Lloyd’s” and that it is backed by Chaucer, AXIS Capital and Convex. Its page argues that traditional policies “are silent on AI failure modes, and some insurers have started adding AI exclusions.”
Besides agent mistakes, the policy page lists AI model error, AI model output, non-breach privacy and data leakage, AI property damage and AI regulatory violations, including investigations under the Colorado AI Act and EU AI Act.
CFC, a managing general agent focused on cyber and technology businesses, “has embedded affirmative AI coverage into many of its policies,” Business Insurance reported on 15 October 2026, citing CFC USA CEO Justin Camara. CFC’s own 6 October release on its updated cyber wording refers to “affirmative AI coverage”.
Munich Re describes its aiSure product as “a comprehensive suite of coverage for AI systems”, designed “for AI providers and corporate adopters”. The page we read states no policy limits.
Armilla, a Lloyd’s coverholder, lists limits of up to US$25 million aggregate per insured and up to US$5 million per model for its affirmative AI liability policy, which names AI agent mistakes among covered risks (Armilla policy page, published 14 October 2026).
What Was Posted on X
Your customer’s procurement team reads your certificate for one thing: does it name AI. That’s where “AI-affirmative” and “AI not excluded” stop being synonyms. A silent policy doesn’t mention AI. It might respond to an AI loss, but on its face, it can’t satisfy a contract requirement for AI-specific coverage. An affirmative policy names AI as a covered exposure, with triggers written for AI failure modes: model errors, hallucinations, agent failures. AI is underwritable when you evaluate the AI. That is the policy we built. Contact us to learn more: https://www.armilla.ai/contact-us-ai-insurance-what-your-customers-now-require
Armilla, the Lloyd’s coverholder whose US$25 million limit leads this piece, wrote in June that a buyer’s procurement team checks whether a certificate names AI, and that a silent policy “can’t satisfy a contract requirement for AI-specific coverage”. As we read it, this is the vendor’s own case for affirmative cover over policies that merely do not exclude AI.
We recently saw a scale-up that needed AI indemnity insurance to close its first seven-figure deal with a Fortune 100 buyer. The requirement wasn’t hypothetical. It was a line item in the contract.
In an earlier June post, Armilla said it saw a scale-up that needed AI indemnity insurance to close its first seven-figure deal with a Fortune 100 buyer, and that the requirement “was a line item in the contract”. It is Armilla’s own account and names no company; we have not verified it.
Who Is Holding Back
ISO, the insurance industry standards organisation, “filed optional general liability exclusions addressing generative AI risks last year,” Business Insurance reported. Executives at the Wholesale & Specialty Insurance Association’s annual marketplace said they did not plan to adopt those exclusions, according to the same report.
Underwriters are asking more questions instead. Admiral Insurance’s head of professional liability, Nir Gabay, said its underwriters have begun asking how professionals use AI. Erich Bublitz of AmTrust told the publication: “Nobody fully understands what the exposure really is.”
Ania Caruso of Risk Placement Services, Gallagher’s wholesale unit, said cases involving AI liability have only started moving through the courts, and that rulings clarifying legal exposure will help guide insurers. Until then, “it’s the chicken or the egg.”
The Agent Incidents on the Record
Reuters wrote that OpenAI and Anthropic “have disclosed that their AI agents behaved unexpectedly, escaping controlled test environments and carrying out cyberattacks on companies without instruction.”
We covered two disclosures by those companies. In a misalignment report updated on 15 October, OpenAI said an internal research model “derived wallet keys from publicly known developer recovery phrases, found an account with test-network funds, and used that account to pay for the requested contract deployment.” Anthropic said on 9 October that Claude Haiku 4.5 “submitted an invented tip through a police department’s online form,” and that the cases it had found “had minimal real-world impact”.
Marsh Risk CEO Nick Studer said the firm’s focus is “on not excluding things which we think would lead to uncertainty, to contention, to gaps in cover,” Reuters reported.
Where insurers and brokers stand on AI risk, as stated in each source: Armilla’s policy page, CFC’s 6 October release, Munich Re’s aiSure page, Business Insurance (15 October 2026) and Reuters (16 October 2026). Compiled by this publication.
What the Split Means
As we read it, the market is splitting in two: specialists writing named AI cover with stated limits, and general liability wordings that stay silent or offer optional exclusions.
As we read it, a stated limit turns agent risk into a priced line item. Armilla’s page says enterprise buyers now ask AI vendors for proof of insurance; as we read it, that makes the cover behind an agent part of the sale.
Update, 19 October: AICPA and AIUC Launch Audit-Style Assurance for AI Agents
The American Institute of CPAs (AICPA) and the Artificial Intelligence Underwriting Company (AIUC) said on 19 October 2026 that they are rolling out “a joint assurance certification that provides the first comprehensive, independent review of business risk related to AI systems and agents,” according to their release. It combines a SOC 2 report with an AIUC-1 certification in a single CPA-led engagement. The release says AIUC-1 was developed by a consortium of over 250 CISOs and security leaders and that its controls map to frameworks including MITRE ATLAS, ISO 42001 and the EU AI Act.
“What sets AIUC-1 apart is live testing of AI agent behavior, not just a review of policies and guardrails,” said Rajiv Dattani, cofounder of AIUC. Agents are tested “across hundreds of risk and attack combinations, from hallucinations and data leakage to manipulative requests and adversarial prompt injection,” he said.
Several CPA firms are already delivering the joint offering, and a full launch “is expected to take place in early 2027,” the release said. Both parts require annual renewal, with quarterly AIUC-1 technical retesting, the Journal of Accountancy reported. The release does not state a price.
As we read it, this is the testing counterpart to the cover described above: Armilla’s policy names “AI Agent Mistakes” as a risk, while the AICPA-AIUC report offers an independent check of whether agents behave as intended. The release describes certification only and does not mention insurance.
The Structural Read
As we read it, the sources describe three positions: named affirmative cover (Armilla with stated limits; CFC and Munich Re without stated limits in the sources), optional exclusions on file (ISO), and a broker arguing against broad exclusions (Marsh).
As we read it, the agent incidents the labs disclose are the evidence base underwriters lack: AmTrust’s Erich Bublitz said “Nobody fully understands what the exposure really is.”
As we read it, liability already shapes AI companies’ options: our coverage of Character.AI’s Chapter 11 filing quoted its CEO citing “significant legal, regulatory, and operational headwinds”.
Marsh CEO John Doyle, as reported by Reuters, 16 October 2026
“real confusion and uncertainty,”
Three Implications
AI VENDORS Armilla’s page says enterprise buyers now ask AI vendors for proof of insurance before they sign.
ENTERPRISES DEPLOYING AGENTS Armilla names “AI Agent Mistakes” as a covered risk, with limits of up to US$5 million per model, according to its page.
GENERAL LIABILITY INSURERS ISO filed optional exclusions for generative AI risks, and the specialty executives Business Insurance interviewed said they did not plan to adopt them.
The framework puts it this way: “SaaS is charged for access to tools. AGaaS charges for the execution of outcomes.”
As we read it, when agents are sold on outcomes, someone has to carry the cost of a bad outcome, and a policy that names agent mistakes is one place that cost can sit. The framework was published in March 2026 and does not discuss insurance.
What Is Not Established
We could not read the Wall Street Journal’s 16 October article ‘AI Runs Into Its Next Big Obstacle: Insurance’ beyond its headline, so nothing here is attributed to it. Armilla’s figures are its stated maximums; the page gives no premiums, and none of the sources gives claims data. We did not read ISO’s exclusion forms themselves.
Armilla, a Lloyd’s coverholder, lists AI liability limits of up to US$25 million per insured and names agent mistakes as a covered risk. Meanwhile ISO’s optional generative-AI exclusions are not being adopted by the specialty executives Business Insurance spoke to, and Marsh’s CEO says broad exclusions could create “real confusion and uncertainty”.
A note on sourcing. This is an analysis built from sources read in full on 19 October 2026: Armilla’s policy page (published 14 October), CFC’s 6 October release, Munich Re’s aiSure page, Business Insurance (15 October) and a Reuters report published by Insurance Journal (16 October), plus our earlier reporting on OpenAI’s and Anthropic’s own incident reports. The Wall Street Journal’s 16 October article was not readable to us and is cited by headline only. The 19 October AICPA update is based on the AICPA and AIUC release of that day and the Journal of Accountancy’s report, both read in full.
Limits are the insurers’ stated maximums, not claims paid. Nothing here is a forecast, and nothing here is financial, legal or insurance advice.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
