Corrections. Every material one, in public.

The matrix said HSB's AI liability cover had no publicly disclosed limit. HSB's release of 18 March 2026 states standard limits of USD 25,000 or USD 50,000 with a USD 500 deductible, and higher limits available. The coverage description is now HSB's own: bodily injury, property damage, and personal and advertising injury arising from the insured's use of AI. The exclusions field now says what the release says, that AI model errors, biased outputs and AI-facilitated data loss are outside it.

Where. The carrier matrix

Verified at munichre.com/hsb

The matrix listed ISO/IEC 42001 and the NIST AI RMF as certifications Munich Re aiSure references, the NIST AI RMF and AIUC-1 for Armilla, and the EU AI Act among AIUC's references. Munich Re and Armilla publish none of these as underwriting references, and AIUC-1 operationalises frameworks rather than the Act. All have been removed. AIUC keeps the frameworks AIUC-1 is published as building on.

Where. The carrier matrix

Verified at munichre.com, armilla.ai, aiuc.com

Several carrier rows carried detailed coverage and exclusion lists that are not on the carriers' own pages: algorithmic bias and privacy failures for aiSure, a nine-item list for Armilla, hallucination and property damage items for Testudo, five worked scenarios for Counterpart, an integrated platform list for Corgi and an eight-item list for Vouch. Each row now states only what the carrier publishes: aiSure settles performance shortfalls on measurable data; Armilla covers AI regulatory violations, including defence costs and insurable fines under the EU AI Act and the Colorado AI Act; Counterpart adds affirmative AI coverage and a technology E&O insuring agreement across miscellaneous professional liability and allied health; Vouch names hallucinations, algorithmic bias and certain intellectual property disputes. A line on larger bespoke aiSure programmes and a description of Chaucer as Armilla's lead syndicate were removed for the same reason.

Where. The carrier matrix

Verified at munichre.com, mosaicinsurance.com, armilla.ai, testudo.co, knowledge.yourcounterpart.com, corgi.insure, vouch.us

The matrix described AIUC's product as certification plus enterprise AI insurance, sold direct, with a specialty indemnity posture. The Artificial Intelligence Underwriting Company is an underwriting company and standards body. Its row now describes AIUC-1, the standard, and the first AIUC-1-backed AI agent policy, for ElevenLabs in February 2026, placed through Lloyd's of London with no carrier named at source.

Where. The carrier matrix

Verified at aiuc.com

Entry c-2026-08-17-10 below removed a 30 April 2025 product launch date for Armilla as unconfirmed. Armilla's own announcement of the launch of its Affirmative AI Liability Insurance with the Lloyd's underwriter Chaucer carries that date on its face, and states that the policy is underwritten by certain underwriters at Lloyd's, including Chaucer. The date is restored to the matrix. The earlier entry is left standing as written.

Where. The carrier matrix

Verified at armilla.ai

The table of carriers tightening AI cover stated that Beazley had confirmed AI sublimit wording in development; beazley.com publishes no AI sublimit wording, and the statement is removed. It listed QBE among the retreating; QBE North America publishes two affirmative AI-focused cyber coverages, an AI regulatory compliance endorsement and LLMjacking protection, and the row now says so. The positions given for W.R. Berkley, AIG, Great American and Chubb rest on trade reporting and could not be read at the carriers' own domains, so the detail is no longer stated and each row says that.

Where. The carrier matrix

Verified at beazley.com, qbe.com

The sourcing note on the ISO generative AI exclusions said none of the forms had been confirmed. Two have since been read from specimens of the forms that carry ISO's own copyright line: CG 40 48 01 26, which removes personal and advertising injury arising out of generative AI, and CG 35 08 01 26, which removes bodily injury and property damage arising out of generative AI from the products and completed operations coverage part. A specimen is one step short of a filing record, and the note says so. CG 40 47 has not been read, and its title and edition are no longer stated, nor is an effective date for any of the three.

Where. The carrier matrix

Read from specimens held at assets.alm.com

A 2024 launch year for Vouch and a July 2025 launch date for Corgi are not published by either company and have been removed. A source line naming Hiscox alongside Vouch has been removed: no acquisition is published. The Counterpart source line gave 21 November 2025; the announcement is dated 24 November 2025.

Where. The carrier matrix

Verified at vouch.us, corgi.insure, knowledge.yourcounterpart.com

The homepage announced a coverage window opening in Q3 2026 and invited registrants into binding quotation; no such window opened, and Agent Insured does not arrange or quote insurance, so both are gone. A market readiness index scored at 34 of 100 had no published method and has been removed, as has an indicative premium framework with an Agent Certified discount that no insurer has committed to. The homepage summary said the Article 5 prohibitions and the Article 4 literacy duty were untouched by the AI Omnibus; the Omnibus added a prohibition and amended Article 4, and the statement is removed. The coverage page cited policy schedules and forms, among them aiSure schedules, an Armilla policy form and a draft Lloyd's AI endorsement, that could not be found at source; the citations are removed. The Monitor page carried a sample issue whose four stories, among them an aiSure schedule addendum and a French court filing, could not be confirmed at any source; it has been removed.

Where. Homepage, coverage, the Monitor, the waitlist

Verified at digital-strategy.ec.europa.eu, munichre.com, armilla.ai, lmalloyds.com

This is the largest correction on this page, and it supersedes entry c-2026-08-15-13 below, which is left standing as written because a log that gets rewritten is not a log. That entry recorded that the Digital Omnibus deferral had not been formally adopted before 2 August 2026 and that the EU AI Act high-risk obligations for Annex III systems had therefore entered into application on that date. That was wrong. The AI Omnibus entered into force on 27 July 2026, six days before the deadline. Annex III high-risk obligations now apply from 2 December 2027 and Annex I obligations from 2 August 2028. What did apply on 2 August 2026 is the transparency layer including Article 50, the governance rules, and enforcement responsibility, which the AI Office and the national authorities have held since that date. The site-wide banner, which carried a superseded "trilogue 28 April 2026" line on 46 pages and four other incompatible variants across the rest, was replaced with one current form. An article whose whole thesis was that the deferral had failed was rewritten around what actually happened and carries its own correction note at the top.

Where. 58 pages, including the site-wide banner, the market tracker, the complete market guide and the 2 August enforcement article

Verified at digital-strategy.ec.europa.eu

Both pages stated that the Artificial Intelligence Underwriting Company projects the AI insurance category at near zero in 2023 growing toward USD 500 billion by 2030, one adding that the estimate is "cited by multiple Lloyd's market briefings". No market-size figure of any kind is published anywhere on aiuc.com. Both statements have been deleted. The directional argument underneath them survives without a forecast, because it can be counted rather than estimated: the number of carriers willing to put an affirmative AI trigger in writing was zero in 2023.

Where. Coalition affirmative AI coverage and Counterpart affirmative AI coverage

Verified at aiuc.com

This site stated that Beazley was developing AI sublimit language capping AI-related payouts at approximately 10 percent of policy limits, and attributed confirmation of that wording to a named head of cyber underwriting management in April 2026. It also stated that draft QBE wording restricts LLMjacking losses to approximately GBP 188,000 on a policy with GBP 3.8 million of limits. Both rest on trade reporting. Neither is published at beazley.com or qbe.com. What QBE does publish points the other way: LLMjacking Protection as an affirmative coverage grant inside an AI-focused cyber endorsement, responding to the service fees an attacker runs up on a hijacked model and the cost of retraining it. Every percentage, the sterling figures and the named individual have been removed across all four pages. The structural observation, that the London market is moving toward sublimits rather than blanket exclusions, is retained and labelled as reported market direction.

Where. The 2026 E&O bifurcation, Lloyd's London market capacity, the market map and the carrier matrix

Verified at beazley.com, qbe.com

Four passages referred to an EIOPA consultation on AI use in the European insurance sector closing in April 2026, and built expectations of forthcoming guidance on it. No AI consultation appears among EIOPA's current consultations and the closed-consultations page does not resolve. All four references have been removed and replaced with the two EIOPA facts this desk can verify: the Opinion on AI governance and risk management of 6 August 2025, and the survey on generative AI of 2 February 2026.

Where. Four pages, including pricing benchmarks, the market map and the carrier matrix

Verified at eiopa.europa.eu

The article stated that the Lloyd's market wrote USD 63.7 billion of gross written premium in 2024 and cited the Lloyd's Annual Report for it. Lloyd's publishes GBP 55.5 billion, against GBP 52.1 billion in 2023, at a combined ratio of 86.9 percent and a profit before tax of GBP 9.6 billion. The USD figure is a conversion Lloyd's does not publish, presented as a Lloyd's figure. It has been replaced with the published numbers and the source link.

Where. Lloyd's London market capacity

Verified at lloyds.com

Corgi was described as the "first full-stack AI-native licensed insurance carrier for startups". corgi.insure states that Corgi Insurance Services, Inc. is a licensed insurance producer and acts as the programme administrator, not the insurer. On pages whose readers are deciding who carries their risk, that distinction is the whole point, and it has been corrected. An ARR figure above USD 40 million, a named investor list and a July 2025 regulatory approval date, all sourced to trade press and one of them to Wikipedia, are not published by Corgi and have been removed. The USD 108 million fundraise is announced on Corgi's own site and is retained.

Where. The market map and the carrier matrix

Verified at corgi.insure

The market map stated that Agent Certified, the certification framework published by Future Proof Intelligence, which also publishes this desk, is "referenced as third-party validation evidence in Munich Re aiSure underwriting documentation". The carrier matrix listed Agent Certified among the certifications Munich Re references. Munich Re publishes nothing of the kind. This is the only self-serving fabrication found in this pass and it is recorded separately for that reason: it asserted that a named reinsurer endorses our own product. Both statements have been removed, and the market map now says plainly that no carrier has adopted Agent Certified as a named underwriting reference.

Where. The market map and the carrier matrix

Verified at munichre.com

The article listed AIUC-1 as covering six dimensions: safety, reliability, alignment, interpretability, robustness and governance. The real six, published by AIUC, are data and privacy, security, safety, reliability, accountability and societal risks. Four of the six stated were invented, and the wrong list appeared in a key takeaway, in the body, in the on-page FAQ and in the FAQ structured data, which means it was being served to AI assistants in machine-readable form. The standard's actual shape, 51 requirements and 130 controls across those six pillars, is now stated. The description of AIUC-1 as "the first published AI insurance certification standard" is also removed: no first-in-market claim is published.

Where. What AI insurance underwriters ask

Verified at aiuc.com

Both articles cited, in their reference lists, a "Lloyd's of London, Emerging Risk Group, AI underwriting framework for managing agents, consultation draft, Q1 2026". Neither could be found at lloyds.com or lmalloyds.com. A reference list is the most authoritative-looking place a claim can sit, which is exactly why an unverifiable one has to come out. Both citations have been removed and replaced with a note saying so. The market-structure description they supported, the coverholder and direct-broker routes into Lloyd's capacity, is verifiable and is retained.

Where. What AI insurance underwriters ask and disclosing AI agents when applying

Verified at lloyds.com, lmalloyds.com

A 30 April 2025 product launch date and a January 2026 capacity announcement date, both from trade press, could not be confirmed at armilla.ai and were removed. Two pages used the form "USD 25 million or more per organisation"; armilla.ai publishes up to USD 25 million per organisation, and "or more" is a different claim. Corrected. The pages now also state directly what the trade press repeatedly gets wrong and this desk inherited: the USD 25 million is a policy limit. Armilla did not raise it. (The 30 April 2025 date was later read on Armilla's own announcement page and restored; see c-2026-10-01-05 above.)

Where. Armilla and the Lloyd's coverholder model, Lloyd's capacity, the market map and the carrier matrix

Verified at armilla.ai

CG 40 47, CG 40 48 and CG 35 08, their titles, their edition dates and the January 2026 effective date attributed to them, come from trade press and law-firm analysis of the ISO multistate general liability filing. None could be confirmed at ISO, at Verisk, or through a state insurance filing portal, most recently on 17 August 2026. The numbers are kept, because they are how a reader finds the endorsement on their own schedule, and every page carrying them now also carries a dated sourcing note that says plainly what they rest on and warns that a search engine confirming them may be drawing on this network's own pages. Two related things were corrected in the same pass. One page dated the forms to December 2023 and gave them titles no other page used, so this site contradicted itself on both edition and title; neither is now stated as fact anywhere. And every sentence that stated the effective date flatly now says "reported". (Two of the three forms were later read from specimens; see c-2026-10-01-07 above.)

Where. Six pages, including the carrier matrix and the three-line coverage framework

Not confirmed at the time. See the sourcing notes on each page.

This is a structural correction rather than a factual one, but it belongs here because it had the same effect. The matrix rendered its whole table in JavaScript, so the raw HTML fetched by a crawler or an AI assistant contained an empty table body. The most citable asset on this site was invisible to the readers it most needed to reach. The rows are now pre-rendered into the HTML, a machine-readable companion is published at carriers.json carrying the same records with per-carrier source URLs and verification dates, and the page declares Dataset structured data. The companion also carries a notVerified block naming what this desk removed and why, so that what we do not know is legible too.

Where. The carrier comparison matrix

Four pages cited LMA5566 as an artificial intelligence exclusion clause, and one described an LMA5566 model endorsement providing a framework for how AI losses may be treated. One reference list carried it as LMA5566 Model AI Endorsement, London, Lloyd's Market Association, 2024, in full bibliographic form. LMA5566 is a state-backed cyber war clause. The current version states 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 no AI subject. The Lloyd's Market Association's own published position is close to the opposite of what this site printed: it treats AI as already caught by the existing definition of a computer system, and describes AI-specific clause work as something still ahead. Every instance has been deleted. The invented reference-list entries were removed rather than retitled, because there is no correct version of a citation to a document that does not exist.

Where. PI or cyber: the AI coverage decision guide, AI agent claims triggers, Sublimits and aggregate caps, Product Liability Directive coverage readiness

Verified at lmalloyds.com

The article described the cyber market's answer to correlated catastrophe loss as the Lloyd's Market Association's systemic cyber catastrophe exclusion clauses LMA5401 to LMA5403, effective from March 2023, and repeated that in its FAQ, its structured data and its reference list: nine instances in all. The story is right and the identification is wrong. LMA5400 to LMA5403 are model cyber clauses for entirely different classes of business, among them Difference in Conditions, Engineering, Nuclear, Power Generation, Cargo, Energy, Fine Art and Marine. The March 2023 requirement came from a Lloyd's market bulletin covering standalone cyber policies, and the model clauses answering it are a different family. Rather than substitute one set of numbers for another, the clause numbers and the effective date have been removed, and the analogy now stands on what the market required, which is what the argument rested on.

Where. Does AI insurance cover systemic model failure and aggregation risk

Verified at lmalloyds.com, assets.lloyds.com

The tracker stated that EIOPA's February 2026 survey found 73 percent of respondents identifying AI governance as a material operational risk concern, up from 51 percent in an equivalent 2024 survey. Neither figure exists and there was no equivalent 2024 survey. The paragraph has been rewritten around the survey's actual findings, published 2 February 2026 across 347 undertakings in 25 countries: nearly two thirds actively using generative AI, and 49 percent with a dedicated AI policy in place, up from about 25 percent in 2023. Because the real figures measure insurers governing their own use of AI rather than enterprise demand for cover, the argument built on the invented number was replaced rather than repointed.

Where. European AI agent insurance market tracker

Verified at eiopa.europa.eu

Several pages described a USD 25 million figure as a January 2026 funding round, variously a Series A and a Series B. Armilla's own announcement describes USD 25 million as the limit of its Standalone AI Liability Policy, up to that amount per organisation. Every reference to a funding round, a capital raise or a series designation has been removed. The limit is retained and now reads as a policy limit.

Where. Multiple articles

Verified at armilla.ai

Pages variously described Armilla's capacity as backed by Chaucer and Axis Capital syndicates, by Chaucer's Lloyd's syndicates 1084 and 1176, and by Chaucer Syndicate. Armilla publishes that Armilla Insurance Services is a Coverholder at Lloyd's and that its Affirmative AI Liability Insurance is underwritten by certain underwriters at Lloyd's, with the product launched with the Lloyd's underwriter Chaucer. No syndicate numbers are published by either party. All syndicate numbers have been removed and the attribution normalised to the published form.

Where. Multiple articles

Verified at armilla.ai

Capacity for the Munich Re aiSure programme distributed by Mosaic Insurance appeared as USD 15 million on some pages and EUR 15 million on others. Mosaic's announcement of 26 February 2026 states up to EUR/USD/CAD 15 million in initial capacity. All references now use that form. The stated audience, AI developers and vendors, has also been restored where pages had widened it to enterprise deployers generally.

Where. Multiple articles

Verified at mosaicinsurance.com

The article described Coalition's endorsement as naming three categories of covered peril: AI-generated output failures, autonomous AI decision errors, and AI-driven data manipulation. Coalition's published endorsement, dated 26 March 2024 for US Surplus and Canada cyber policies, does two narrower things: it expands the security failure and data breach definition to include an AI security event, and it expands the funds transfer fraud trigger to cover a fraudulent instruction transmitted through deepfakes or other AI technology. The description has been corrected to the published scope, and the analysis now says plainly that the endorsement does not respond to loss caused by a wrong AI output.

Where. Coalition affirmative AI coverage: European relevance

Verified at coalitioninc.com

The market tracker attributed Counterpart's affirmative AI coverage to management liability and directors and officers lines. Counterpart's November 2025 announcement expanded its Affirmative AI Coverage and added a technology errors and omissions insuring agreement across its miscellaneous professional liability and allied health products. The line attribution has been corrected.

Where. Multiple articles

Verified at the Counterpart announcement, 24 November 2025

Pages described HSB's small-business AI liability product as launched in March 2026 with a simplified underwriting process, and set out documentation requirements for it. HSB's announcement of 18 March 2026 states that the cover is added to the business policies of carriers that partner with HSB rather than sold direct to businesses, and that availability is pending insurance regulatory approval. The invented underwriting description has been removed and the distribution model stated as published.

Where. Multiple articles

Verified at munichre.com/hsb

Testudo was described as launching in January 2026 with Apollo, Atrium and QBE capacity. Testudo launched on 21 January 2026 with Apollo capacity; Atrium and QBE joined on 26 February 2026, lifting the limit to USD 9.25 million per insured. The product is written for United States enterprises, which pages had not made clear. Atrium's Lloyd's syndicate number, which neither party publishes in this context, has been removed.

Where. Multiple articles

Verified at testudo.co, atrium-uw.com

A stamp capacity growth figure, a capacity threshold in pounds, a corporate ownership attribution and two financial strength ratings were stated for Chaucer's syndicates. None could be verified at Chaucer's own domain. All have been removed. The qualitative statement of Chaucer's role as lead capacity provider on the Armilla binding authority is retained.

Where. The Lloyd's market and AI liability

Verified at chaucergroup.com

The toolkit was dated 24 April 2026. The Lloyd's Market Association published it on 23 April 2026. The date has been corrected. The survey findings cited alongside it, 93 percent of respondents with or developing formal AI frameworks across firms representing over 60 percent of Lloyd's stamp capacity, are confirmed at the LMA's own domain and are retained.

Where. The Lloyd's market and AI liability

Verified at lmalloyds.com

AIUC appeared expanded as the AI Underwriting Consortium, the AI Underwriter Collective and the AI Underwriting Company. The organisation is the Artificial Intelligence Underwriting Company. Every expansion has been normalised, and a link to a non-resolving domain has been corrected to aiuc.com.

Where. Multiple articles

Verified at aiuc.com

The EIOPA opinion on artificial intelligence governance and risk management was cited with two different document reference numbers on different pages. Neither could be verified. The reference numbers have been removed. The opinion and its date of 6 August 2025 are retained.

Where. Multiple articles

Verified at eiopa.europa.eu

Several pages still described the Digital Omnibus deferral of the EU AI Act high-risk deadline as a live proposal that might yet move the 2 August 2026 date. The deferral was not formally adopted or published in the Official Journal before that date, so the original date governed and the high-risk obligations for Annex III systems entered into application on 2 August 2026. The pages were updated to that position.

Where. Multiple articles

Verified at Regulation (EU) 2024/1689

Descriptions of named carriers' internal underwriting practice, submission questionnaires, documentation checklists, premium floors and forward-looking market plans were stated as fact in several places without a published source. Where a source could not be found at the company's own domain, the specific description has been removed and replaced with reasoning from the product's published structure, which is verifiable and does not put words in a carrier's mouth.

Where. Multiple articles

This site described Munich Re aiSure as a parametric product, and in places as a parametric category of insurance in its own right. Munich Re does not use that word for aiSure anywhere it publishes. What Munich Re and Mosaic Insurance do publish is that aiSure is a performance guarantee for AI systems that settles on measurable performance data, and that Munich Re requires its own technical due diligence before cover is written. Those are different claims: a parametric trigger pays on an index without asking about the loss, which is a specific and consequential product design. The label was ours, and it has been removed from every description of aiSure and of Munich Re. Parametric remains in use on this site where it describes the insurance model generally rather than this product, because that is ordinary industry vocabulary and accurate.

Where. 45 pages, including the aiSure article, the market map, the pricing benchmarks and the carrier comparison tool

Verified at munichre.com, mosaicinsurance.com

The aiSure article was published at a URL ending munich-re-aisure-parametric-ai-insurance-europe. Correcting the body copy left the assertion standing in the address, in the page's structured data, in the sitemap and in the machine-readable index this site serves to AI assistants. A slug is an identifier rather than a sentence, but this one was being read as a sentence by the systems we publish for. The article now lives at munich-re-aisure-ai-performance-insurance-europe and the old address issues a permanent redirect to it. The title and the structured data have been changed to match.

Where. Munich Re aiSure: AI performance insurance explained

Verified at munichre.com

aiSure was called the world's first parametric AI performance insurance product in page titles, meta descriptions, structured data and the feed. Nobody publishes that, and a first-in-market claim about a named reinsurer's product is exactly the kind of statement that should never rest on our own enthusiasm. Every instance has been deleted rather than qualified. The same pass reduced world's first Lloyd's coverholder for Armilla to first Lloyd's coverholder dedicated exclusively to AI liability, which is the form Armilla itself publishes.

Where. The aiSure article, the market topic index and the site feed

Verified at munichre.com, armilla.ai

Pages stated that Munich Re began writing AI performance risk in 2018, that aiSure had covered large language models since 2019, and that the product sits in a Munich Re Special Enterprise Risks division. None of that is published by Munich Re. All three have been removed. One page also dated the aiSure and Mosaic partnership to 2024 or 2025; the announcement is 26 February 2026.

Where. The aiSure article and several reference lists

Verified at munichre.com, mosaicinsurance.com

Residual statements attributed the first AIUC-1-backed policy to Munich Re, in one case saying Munich Re issued the policy on the strength of the certification. Neither the Artificial Intelligence Underwriting Company nor ElevenLabs names an insurer or a reinsurer for that policy, and no limits are disclosed. Every carrier attribution has been removed. What is published, and retained, is that the policy was placed through Lloyd's of London.

Where. Homepage, and several articles

Verified at aiuc.com, elevenlabs.io

Four further expansions of AIUC survived the first pass in footnotes and reference lists: AI Underwriting Council, AI Underwriting and Certification Consortium, The American AI Underwriting Collaborative, and a bare AI Underwriting Company. The company is the Artificial Intelligence Underwriting Company and nothing else. One of those entries also dated the AIUC seed round to late 2024; it was July 2025, led by Nat Friedman at NFDG.

Where. Reference lists on several articles

Verified at aiuc.com

Three dates were checked against the parties themselves. Testudo's capacity expansion with Atrium and QBE was given as March 2026 in five places; Testudo dates it 26 February 2026 and every instance now reads that way. The LMA AI Adoption Toolkit appeared as both 23 and 24 April 2026; the Lloyd's Market Association publishes 23 April 2026. Armilla's USD 25 million was in two places still cited to a trade report headlined as a raise. The limit is verified at armilla.ai and retained; the trade citation is replaced with Armilla's own page. Armilla's announcement page carries two conflicting dates for that limit, one year apart, so this site states the limit without a date.

Where. Lloyd's capacity article, market map, carrier comparison tool

Verified at testudo.co, lmalloyds.com, armilla.ai

Hamilton Insurance Group was described as having sought regulatory clearance for AI exclusions in management liability policies. Other trade reporting describes Hamilton's position instead as a sublimit inside errors and omissions or cyber cover, which is the opposite outcome for a buyer: one removes the cover, the other caps it. Hamilton publishes nothing about AI underwriting on its own site, so neither version can be checked at the carrier. This site now states neither and says so on the page.

Where. Market map, carrier comparison tool

No statement at hamiltongroup.com