Future Proof The Authority Stack
Insure Your Agent

Corrections

A log of material corrections made to anything published on this site. Our Editorial Standards commit us to maintaining it. Readers who believe they have identified an error are invited to write to operators@futureproofintelligence.com.

The Editors Last updated 15 August 2026

How we handle corrections

A material correction is one that changes the substance of a claim: a figure, a date, a named party, a legal effect, or the availability of a product. Material corrections are logged on this page with the date, the page affected, what was wrong, and what it now says. The original wording is quoted so the record is auditable.

A non-material correction is a typographical fix, a broken link, or a formatting change that does not alter meaning. Those are made silently and are not logged here.

We do not delete pages to resolve errors. We correct them in place and record the correction.

Log

15 August 2026: editorial review of third-party claims

A full review of every claim on this site naming a third-party insurer, carrier, underwriter, standards body or regulator. Any claim that could not be verified against the named organisation's own published material was removed rather than reworded. The following material corrections were made.

  • Armilla coverage limits. Multiple pages stated that Armilla offered limits of up to USD 25 million, in several cases attributing that figure to a January 2026 funding round. Armilla publishes no standard limits. Those figures have been removed. Armilla's status as a Lloyd's coverholder and managing general agent, and its capacity providers Chaucer Group, Axis Capital and Convex, are published by Armilla and have been retained.
  • Armilla funding. Two pages described a USD 25 million round raised from Chaucer and Axis Capital, one of them calling it a Series A. Chaucer and Axis Capital are Armilla's insurance capacity providers, not investors in that round, and Armilla publishes no funding figure. The claim has been removed.
  • Testudo capacity and underwriters. Several pages stated capacity of USD 9.25 million per insured, named Apollo, Atrium and QBE as underwriters, and dated the launch to January 2026. Testudo publishes none of those figures and states that its cover is underwritten at Lloyd's, with Testudo UK Limited acting as an Appointed Representative of Pro MGA Solutions Ltd, regulated by the Financial Conduct Authority. The unverified figures have been removed and replaced with what Testudo publishes.
  • HSB pricing. Two pages stated pricing of approximately USD 500 to 2,500 per year for limits of USD 250,000 to 1,000,000. HSB disclosed no pricing, deductibles or limits at launch. The figures have been removed. The coverage categories, the 18 March 2026 launch date, the distribution model through partner carriers, and the survey figures of 74 per cent and 91 per cent are stated in HSB's own launch release and have been retained.
  • HSB product rating. Our review of the HSB product carried a machine-readable star rating of 4 out of 5. We are not a licensed insurance intermediary and it is not appropriate for us to publish a structured score of a regulated financial product. The rating has been deleted. The analysis of what the product covers and does not cover remains, because that is the work this site exists to do.
  • HSB illustrative limits and deductibles. One section presented ranges for limits, deductibles and premium described in an editorial note as illustrative rather than actual. Illustrative figures attached to a named regulated product invite exactly the misreading the note was trying to prevent. They have been removed.
  • ElevenLabs AIUC-1 policy. Several pages stated the policy was underwritten with Munich Re reinsurance participation. AIUC and ElevenLabs describe the coverage as placed through Lloyd's of London and name no reinsurer. The Munich Re attribution has been removed.
  • Munich Re aiSure availability date. Several pages stated the product had been available since 2018 and one added that it extended to large language model coverage in 2019. Munich Re publishes neither date. Both have been removed. The description of aiSure as a performance guarantee with parametric-like settlement is Munich Re's own and has been retained.
  • AIUC name. AIUC was expanded as the AI Underwriting Company. Its registered name is the Artificial Intelligence Underwriting Company. Corrected throughout. The link to aiuc.co on the homepage did not resolve and now points to aiuc.com.
  • Agensure. Our entry stated that Agensure referenced EU AI Act Article 50 and that insurance was on a Year 2 roadmap through an MGA structure. Agensure maps its certificate to Article 15 on accuracy, robustness and cybersecurity, and publishes nothing about insurance or a roadmap. Both claims have been corrected. The Agensure Risk Score, scaled 1 to 100, and the Agent Deployment Readiness certificate are as published.
  • Klaimee and Mount carrier status. Our entries described Klaimee as possibly a licensed carrier or MGA and Mount as the first AI agent liability carrier. Neither company publishes its underwriting structure or its underwriter. Both entries now say so rather than estimating. Mount's description of itself as an AI agent insurance carrier is presented as Mount's own claim.
  • Chubb AI exclusion. A section described Chubb as carving out systemic AI events while retaining cover for isolated incidents. The single source supporting it is no longer online and the wording could not be verified against Chubb's own material. The section has been removed.
  • Market pricing ranges. Homepage and article answers gave euro and dollar ranges for cyber, technology errors and omissions, and standalone AI liability premiums. No carrier in this market publishes a rate card and those ranges were estimates. They have been replaced with a plain statement that pricing comes from a broker submission.
  • Unresolved verification markers. Seventeen editorial placeholders reading "VERIFY" were live in the published body copy of our European market reference page. Each has been resolved: the claim beside it was either confirmed against the provider's own material or removed.

15 August 2026: second pass on Munich Re, HSB and AIUC

A later review the same day found a further set of claims about named third parties. These corrections supersede anything above that conflicts with them.

  • Munich Re aiSure described as parametric. The entry above retained the phrase performance guarantee with parametric-like settlement on the basis that it was Munich Re's own. It is not. Munich Re does not use the word parametric for aiSure anywhere it publishes. What Munich Re and Mosaic Insurance publish is that aiSure is a performance guarantee for AI systems settling on measurable performance data, with Munich Re technical due diligence required before cover is written. The word has been removed from every description of aiSure and of Munich Re across 24 pages, including the homepage, the coverage audit tool and the SME product comparison. It remains where it describes the insurance model generally, which is ordinary industry vocabulary and accurate.
  • Munich Re Special Enterprise Risks. Three pages placed aiSure in a Munich Re Special Enterprise Risks division and one said the product is generally placed through it. No such divisional attribution is published. All have been removed.
  • HSB as first in market. Five statements described the HSB product as the first tier-one reinsurer-backed SME AI liability cover, the first purpose-built product of its kind, or the first mass-market SME AI liability product. HSB's own launch release makes no first-in-market claim of any kind. Every such claim has been deleted. The 18 March 2026 launch date, the coverage categories and the distribution model through partner carriers are HSB's own and are retained.
  • AIUC name, second pass. Seven further instances expanded AIUC as the AI Underwriting Consortium or as a bare AI Underwriting Company. The company is the Artificial Intelligence Underwriting Company. Corrected throughout.
  • Munich Re participation in AIUC-1. The homepage, the get covered page and one article stated that AIUC-1 was developed with Munich Re, or with Munich Re participation, and one presented that as a case file datum. AIUC publishes AIUC-1 as its own standard and names no insurer or reinsurer for the ElevenLabs policy written against it. Every attribution has been removed.
  • ElevenLabs policy date. The policy was dated 11 February 2026 in two places. Both aiuc.com and elevenlabs.io date the announcement 12 February 2026. Corrected.
  • Hamilton Insurance Group. Our AI exclusion article described a Hamilton generative AI exclusion for professional liability, quoted its trigger wording, and listed the generative AI tools it names. Other trade reporting describes Hamilton's position instead as a sublimit inside errors and omissions or cyber cover, which for a buyer is the opposite outcome. Hamilton publishes nothing about AI underwriting on its own site, so neither version can be checked at the carrier. The quoted wording and the line-of-business characterisation have been removed and the page now says why. The coverage audit tool no longer names Hamilton alongside AIG.

17 August 2026: London market clause numbers, and the sourcing of the ISO form numbers

The Lloyd's Market Association publishes part of its clause library without a login, so the clause numbers this site cited could be read rather than assumed. Two did not survive. Both were the same shape of error: a real clause number issued by a real body, printed here with a subject that is not its own. That is harder to catch than an invented name, because the number checks out the moment anyone looks it up and only the meaning is wrong.

  • LMA5566 described as an AI exclusion. The pre-deployment checklist and the broker conversation guide both said the Lloyd's Market Association introduced AI exclusion language, LMA5566, in 2023, and both carried it in their reference lists, one of them as LMA5566: Artificial Intelligence Exclusion (2023), with an invented description of the losses it addresses. LMA5566 is a state-backed cyber war clause. The current version carries its own title on its face: War and Cyber Operation Exclusion No. 3, for use on commercial cyber insurance contracts, dated 18 January 2023. It has nothing to do with AI, and the association's own published position is close to the opposite of what we printed: it treats AI as already caught by the existing definition of a computer system and describes AI-specific clause work as still ahead. Every instance has been deleted and the two reference entries removed. What both pages actually needed to say, and now say, is that AI-related exclusions and carve-outs are appearing in technology professional indemnity and cyber renewals, which is observable and was never the part that depended on the clause number. Verified at lmalloyds.com.
  • An LMA5400 series of AI model exclusion clauses. A reference on the wrong-advice liability article cited Lloyd's Market Association AI Model Exclusion Clauses LMA5400 series (2024 to 2026). No such series exists. LMA5400 to LMA5403 are model cyber clauses for classes of business including Difference in Conditions, Engineering, Nuclear, Power Generation, Cargo, Energy, Fine Art and Marine, and the association publishes no AI clause series at all. The date range was invented alongside the subject. The citation has been removed rather than renumbered. Verified at lmalloyds.com.
  • The ISO CG form numbers now show their sourcing. This site cites ISO endorsements CG 40 47, CG 40 48 and CG 35 08, and one article is built on them. We tried again on 17 August 2026 to confirm those form numbers and their January 2026 effective date against the filed forms, at ISO, at Verisk and through state filing portals, and could not. They are attested by trade and law-firm analysis only. The numbers have not been removed, because the analysis of what these endorsements do is useful and the numbers are how a reader finds the endorsement on a renewal schedule. Instead the sourcing is now stated wherever they appear, including in the machine-readable file this site serves to AI assistants: reported by named industry sources, not confirmed against the filed forms. One reference also carried a 2023 edition year for CG 40 47, which contradicts the January 2026 date used elsewhere on this site; that year has been removed rather than picked between. A related caution for anyone checking this: a search engine will return a confident answer confirming the January 2026 date, assembled from a cluster of pages that includes this one. Our own page reflected back at us is not verification, and we do not treat it as such.

17 August 2026, later the same day: two of the three ISO forms were read, and a great deal that stood on them did not survive

The entry above concluded that the ISO CG form numbers could not be confirmed and settled for stating their sourcing. That conclusion was half wrong, and this entry supersedes it on the point. Specimens of two of the three forms turned out to be publicly obtainable after all, each marked SAMPLE and footed with the Insurance Services Office copyright line. CG 40 48 01 26 and CG 35 08 01 26 have now been read. CG 40 47 01 26 has not, and no specimen of it was obtainable on 17 August 2026, so every page that names it now says so.

Reading two forms settled a question this site had been answering five different ways. CG 40 48 01 26 is titled Exclusion, Generative Artificial Intelligence (Coverage B Only). It is one page. It adds one sentence of exclusion, for personal and advertising injury arising out of generative artificial intelligence, and one sentence of definition: a machine-based learning system or model trained on data with the ability to create content or responses, including but not limited to text, images, audio, video or code. There is no exception language. CG 35 08 01 26 does the same on the products and completed operations coverage part, for bodily injury and property damage, in identical words. Against that, the following came off the site.

  • Five different titles for one form. CG 40 47 was described here as Exclusion, Access or Disclosure of Confidential or Personal Information and Data-Related Liability; as Exclusion, Recording and Distribution of Material or Information in Violation of Law; as Exclusion, Artificial Intelligence; as Exclusion, Certain Artificial Intelligence; and as Exclusion, Expected or Intended Injury. The first two belong to unrelated ISO endorsements. The others are invented. All have been removed, and no title is now given for a form we have not read.
  • Three different introduction dates. The endorsements were dated variously to 2023, December 2023, 2024 and January 2026 across this site. The two forms we have now read carry the 01 26 edition date and a 2025 copyright line. Every 2023 and 2024 date has been removed.
  • Policy wording we had never seen, quoted as if we had. One article told readers that CG 40 47 excludes loss caused directly or indirectly by any artificial intelligence, and that CG 40 48 preserves cover for AI that does not independently make decisions and does not autonomously operate without human oversight. None of that language is on the forms. The autonomy test in particular is the opposite of what the real definition does, which draws no distinction between an autonomous agent and a member of staff using a mainstream assistant. That article now quotes the real wording, and says which form it was read from.
  • An entire exclusion section resting on the wrong form. Our claims guide taught that CG 40 47 and a W.R. Berkley form both exclude expected or intended outcomes, and built practical advice on it. The ISO generative AI endorsements contain no state-of-mind test at all. The expected-or-intended injury exclusion lives in the standard commercial general liability form, and the parallel restriction in specialist AI wordings is a deliberate-act or knowing-misuse exclusion. The section has been rewritten to say where the exclusion actually comes from. The practical advice was right; the citation was not.
  • The W.R. Berkley form number is gone from this site. We printed a Berkley form number in eight places and described its wording in detail, including a quoted definition of artificial intelligence and a list of trigger phrases. In one place we called it an absolute exclusion for D&O, E&O and fiduciary liability, and in another an affirmative AI coverage endorsement. Those are opposite instruments and both cannot be true. W.R. Berkley publishes no forms library, no forms list and no statement on AI underwriting, so neither version can be checked at the carrier. The number and the wording description have been removed everywhere, including from the coverage audit tool. What remains is the part that is true and useful: management liability and professional lines carriers are filing AI exclusions of their own, they typically define artificial intelligence broadly rather than limiting themselves to generative AI, and the endorsement on your own schedule is the only thing that binds.

17 August 2026: carrier claims, pricing, and the legal sector guide

The same pass reviewed every statement on this site that tells a reader what a policy does or does not cover, on the principle that this is the class of claim a reader will act on at renewal.

  • Beazley and QBE cyber sublimits. The AI exclusion guide stated that both carriers had moved to sublimits capping AI-related cyber losses at approximately 10 per cent of policy limits, gave two sterling figures for the effect on specific policy sizes, and attributed a statement about in-development wording to a named Beazley underwriter. All of that came from trade summaries of newspaper reporting. Neither carrier publishes an AI or LLMjacking sublimit on its own site. The percentages, the figures and the named attribution have been removed. What QBE North America does publish, and what the page now says, is affirmative AI-focused cyber endorsements for AI regulatory compliance and for LLMjacking, with no sublimit stated.
  • Corgi revenue and approval date. Two passages stated that Corgi received full regulatory approval in July 2025 and had exceeded USD 40 million in annual recurring revenue by early 2026, sourced to a magazine summary and a startup directory profile. Corgi publishes neither figure. Both have been removed.
  • Hiscox AI coverage. A passage stated that Hiscox had launched revised technology professional liability policies in 2025 with explicit cover for AI-related claims including model failures and negligent AI advice. Hiscox publishes nothing about AI cover or AI exclusions in that product. The page now reports that the revision has been reported and does not describe the coverage.
  • A carrier carve-out for systemic AI events. A named carrier was again described as granting AI cover while excluding widespread incidents where one model failure reaches many clients. This was removed once on 15 August and had returned in two further passages sourced to a trade summary. It has been removed again, and replaced with the general point, which is sound: systemic-event and aggregation carve-outs are the thing to read for in any AI grant, because systemic exposure is what carriers manage hardest.
  • A litigation growth statistic. A key takeaway stated that generative AI lawsuits in the United States grew 978 per cent between 2021 and 2025. The source was a market update published by a company selling into that market. We do not publish a market-size or growth figure on the strength of a vendor's own marketing. It has been removed and replaced with two verified case outcomes.
  • HSB survey figures and HSB limits, in opposite directions. The HSB review carried survey figures of 74 and 91 per cent that are not in HSB's release; they have been removed, superseding the 15 August entry above which retained them. In the other direction, this site said across four pages that HSB disclosed no limits at launch. It did: standard limits of USD 25,000 or USD 50,000 with a USD 500 deductible, higher limits available. Suppressing a published limit is its own kind of error, and for an SME reader it is the single most decision-relevant fact about the product, so it is now stated prominently. The release also names three exclusions expressly, that the product does not cover AI model errors, biased AI outputs or AI-facilitated data loss, and a reference note here had described the product as covering AI-generated errors and omissions, which is close to the reverse. Corrected.
  • The pricing guide has been rewritten. Our guide to what AI insurance costs was built on euro ranges for cyber endorsements, technology errors and omissions, and standalone AI policies, plus a claim that documented governance produces premiums 15 to 30 per cent lower. Not one of those figures could be sourced to a carrier, and no carrier or managing general agent in this market publishes a rate card. A pricing page built on estimates reads as authoritative and is not, and a reader takes those numbers into a broker meeting. The ranges and the percentage are gone. The page now sets out the limits that are actually published, which turn out to tell the real story about this market, the five things that determine a quote, and a three-week route to a real number instead of a range.
  • Mata v. Avianca, on the page most read by lawyers. Our legal sector guide, and the SME handbook, both named the plaintiff, Roberto Mata, as one of the sanctioned attorneys. The sanctioned parties were Steven A. Schwartz and Peter LoDuca, together with the firm, jointly USD 5,000, by opinion and order of Judge P. Kevin Castel dated 22 June 2023. One other page dated the sanctions order to May 2023 in its body while dating it to June in its own reference. All corrected.
  • SRA and Bar Council citations. The legal sector guide attributed the duty of competence to SRA Principle 4 and the duty of confidentiality to Principle 6, cited a Chapter 6 of the 2019 Code, and relied on an SRA AI guidance document dated 2024. The SRA Principles are seven and include neither competence nor confidentiality: Principle 4 is honesty and Principle 6 is equality, diversity and inclusion. The obligations are in the Code of Conduct for Solicitors, RELs and RFLs at paragraphs 3.2, 3.3 and 6.3, and the 2019 Code has no chapters. The SRA's position on AI is in its Risk Outlook report of 20 November 2023, which the page now quotes directly, including its instruction that a firm should not trust an AI system to judge its own accuracy and its expectation that firms tell clients when AI will be used. The Bar Council document was cited with a title, a year and section numbers that do not match the guidance the Bar Council actually publishes; it is now cited correctly, with the note that the Bar Council states it is not guidance for the purposes of the BSB Handbook. A description of the cab rank rule as the barrister's duty to the court has been corrected: the cab rank rule governs the acceptance of instructions.
  • Underwriting conditions attributed to three providers. The legal sector guide told readers that Armilla's cover responds where the insured can demonstrate reasonable verification steps were taken, that Counterpart's requires a defined verification standard to have been met, and that AIUC-1 certification unlocks terms unavailable to uncertified systems. None of those conditions is published by any of them. They read as underwriting detail and they would change a firm's buying decision. All three have been removed, and the section now states what each provider actually publishes and says plainly that none of these products is built for a firm of ten to two hundred people.
  • Two citations that could not be found at all. The same guide cited a Lloyd's publication, Artificial Intelligence: Underwriting Considerations (2023), and a Chartered Insurance Institute guidance note, AI and Professional Liability: A Guidance Note for Brokers (2025). Neither could be located at lloyds.com or cii.co.uk. A citation that cannot be found is worse than no citation, and both have been removed.
  • Lloyd's Market Bulletin Y5381. A reference described it as a July 2023 bulletin about cyber war exclusions and AI systems. It is dated 16 August 2022, is titled State backed cyber-attack exclusions, required standalone cyber policies in risk codes CY and CZ to carry the exclusion from 31 March 2023, and says nothing about artificial intelligence. Corrected.
  • Reinsurers named for market pressure. Two passages named Munich Re and Swiss Re as having required their cedents to resolve silent AI exposure. Neither publishes its cedent requirements, and repeating a market rumour with two real names attached to it is exactly the shape of error this log exists to record. The names have been removed; the underlying market movement, which is observable, is described without them.
  • A stale Colorado date on the homepage. The homepage listed the Colorado AI Act as a forthcoming 30 June 2026 deadline. That date has passed and the reported effective date moved at least once during 2025, which we could not settle at the Colorado legislature's own pages. The line now says the Act is operative rather than naming a day we cannot confirm.
  • Article 50 of the EU AI Act, in the other direction. Nothing here was wrong, but something useful was missing. Article 50 transparency obligations still apply from 2 August 2026 and were not moved by the AI Omnibus. There is one narrow relief, which this site did not mention anywhere: AI systems already placed on the market before 2 August 2026 have until 2 December 2026 to meet the Article 50(2) obligation to mark AI-generated content. That has been added across the site.

Standing note on schema

Structured data describing a third party's regulated insurance product as something we rate, or describing an insurance product as carrying the Future Proof Intelligence brand, has been removed from this site. We publish analysis and reference material. We do not underwrite, broker, rate or sell insurance, and our machine-readable data now says the same thing our footer does.