Munich Re Chief Executive Officer Christoph Jurecka used the reinsurer's July 8, 2026 investor briefing to lift the wraps on aiSure, a purpose-built cover for losses caused by artificial-intelligence errors, and to confirm that Bermuda-domiciled specialty carrier Mosaic Insurance would ride alongside as the primary distribution partner. The product responds on a parametric basis, uses fixed model-accuracy triggers rather than proof-of-loss adjudication, and caps out at a per-claim limit of €15M, roughly $15M at the July 8, 2026 spot rate (Munich Re, July 2026).
What aiSure Actually Covers
The four peril categories aiSure names in its policy wording are unusually explicit for a specialty cover at launch: large language model hallucinations that produce commercially harmful outputs, algorithmic-discrimination findings arising from automated decisioning, copyright-infringement claims tied to generative-model training or output, and regulatory fines imposed by AI-specific supervisors. The Munich Re press materials also carve out a fifth adjacent category, model-performance shortfalls measured against contractual benchmarks, which Mosaic's original MGA program treats as a separate performance-warranty layer (Mosaic Insurance, February 2026). The product page at Munich Re frames the target buyer as an AI vendor or an enterprise deploying vendor models under a defined service-level agreement (Munich Re, July 2026).
The parametric structure is the load-bearing design choice. Traditional technology errors-and-omissions covers require the insured to demonstrate a claim, prove causation to a defined loss, and negotiate a settlement inside a contested claims process that can run 18 to 36 months. aiSure replaces that adjudication with a fixed trigger: if a defined model-accuracy benchmark falls below a contractual threshold, or if a named regulatory finding is issued against the insured, the cover pays a pre-agreed amount without a subjective loss quantification. Reinsurance News reported that the parametric mechanism was the element Mosaic underwriters spent the longest defending during the product's development cycle, precisely because the parametric approach compresses claims-handling expenses to almost nothing while shifting the underwriter's problem back onto trigger calibration (Reinsurance News, July 2026).
The named senior underwriter for the Mosaic side, Krishnan Ethirajan, framed the design intent in his comments accompanying the February 2026 launch of the MGA program: "AI developers face a rapidly evolving risk landscape, and traditional insurance products are not designed to keep pace. Our collaboration with Munich Re allows us to offer a solution that not only mitigates financial risk but also supports responsible AI deployment." (The Insurer, February 2026). That framing, mitigating financial risk while supporting responsible deployment, is the softer language accompanying the harder underwriting decision: Mosaic and Munich Re are betting they can price a peril class that has almost no loss history in the actuarial sense, using observable performance metrics as the proxy for insured loss.
How the Parametric Trigger Actually Works
The fixed-trigger mechanism inside aiSure is the piece most likely to reshape how actuaries reserve the cover. Traditional professional-liability and tech E&O policies produce a long-tail loss development pattern: notified claims mature slowly, incurred-but-not-reported reserves rely on Bornhuetter-Ferguson or chain-ladder projections applied to sparse triangles, and the ultimate loss ratio for an accident year is not credibly known for five to eight years. Parametric covers collapse that timeline. The trigger fires or it does not, the settlement amount is contractual, and the loss ratio for a policy year crystallizes in months rather than years.
The trade-off is basis risk. The insured's actual economic loss from an LLM hallucination event, or from an algorithmic-discrimination finding, may bear little resemblance to the parametric payout tied to a benchmark-threshold breach. A Munich Re-published overview of the product notes that the benchmark thresholds are calibrated per insured, using a pre-bind model-validation exercise that measures the customer's specific model against a battery of accuracy tests (Munich Re, July 2026). That calibration is the underwriting act. If the underwriter sets the threshold too tightly, the trigger fires on ordinary model performance variance and the cover behaves like a stop-loss on model quality rather than an insurance product. If it is set too loosely, the trigger only fires on catastrophic model failure and the insured perceives the cover as economically empty.
Michael Berger, who heads Munich Re's Insure AI unit and has been the technical face of the product since Munich Re first published its aiSure landing page, has been explicit that the calibration exercise is where the intellectual property sits. In a public statement referenced in Munich Re's product materials, Berger has repeatedly framed the aiSure book as an actuarial rather than a legal proposition: the underwriter's judgment on trigger placement, not the wording of the policy, determines the loss ratio (Munich Re, July 2026). That framing is unfamiliar in specialty casualty but reads naturally as parametric weather or parametric cyber, both of which live in the same design pattern.
The €15M Limit in Context
A €15M per-claim limit is a modest number relative to the model-scale AI losses that have already reached litigation or public settlement. The Air Canada chatbot decision from British Columbia's Civil Resolution Tribunal in February 2024 set the direct-liability precedent at a small dollar figure but opened a much larger channel of secondary claims tied to airline and travel-industry chatbot outputs. Deepfake-fraud losses reported through 2025 and into the first half of 2026 have run into the tens of millions for individual corporate victims, with the widely-cited Arup Group engineering-firm deepfake fraud alone quantified at roughly $25M (Royal Gazette, February 2026). Against that loss universe, €15M covers the mean loss for a small-to-midsize AI vendor but does not begin to touch the tail-scenario losses a large enterprise deployer might face from a single model-failure event.
The implication for placement structure is that aiSure will not stand alone on a serious insured's tower. It will either be layered under a bespoke facultative program the buyer arranges from other markets, or it will operate inside a Mosaic-led syndicated placement where multiple carriers stack their capacity to produce a tower well above the €15M primary limit. Mosaic's original MGA press materials confirmed the design intent of syndicated stacking, noting that the aiSure product would function inside a broader Mosaic-arranged program rather than as a single-carrier standalone (Mosaic Insurance, February 2026). For actuaries at other carriers, the placement question is how to price excess layers above a Munich Re primary that itself uses parametric triggers, because the loss trigger for the excess layer must be defined against the primary's parametric event rather than against a traditional attachment-point exhaustion.
Pricing a Peril With Almost No Loss History
The pricing problem aiSure sits inside is the same problem that dominates the AI-liability coverage thin-data pricing methodology discussion actuaries have been running since 2024. Munich Re has one of the deepest specialty-lines reserving datasets in the industry but no prior book of parametric AI-error losses to feed a traditional loss-cost projection. The workable analogs are three: the emergence of cyber as a specialty class starting around 2015, the parametric weather-cover pricing that reinsurers refined during the 2010s catastrophe-modeling build-out, and the professional-liability tail loss patterns from technology errors-and-omissions programs written for hyperscale cloud vendors from 2018 through 2024.
None of the three is a perfect fit. Cyber circa 2015 had the same absence of loss history but relied on subjective proof-of-loss claims adjustments, which produced a decade of adverse development as insurers underestimated ransomware and business-email compromise severities. Parametric weather has the same trigger design but relies on decades of physical-observation data to calibrate thresholds. Tech E&O carries the same insured universe (AI vendors substantially overlap with cloud and software vendors) but the loss triggers and adjudication patterns are different enough that direct pattern transfer is limited. What Munich Re gains from the parametric design is that the trigger produces a directly observable claim-frequency signal within months of policy inception, so credibility on the pricing assumption builds faster than it did for cyber. What Munich Re loses is that mispriced parametric triggers cannot be quietly adjusted through the claims process the way subjective loss picks can be tightened; the calibration decision is committed at bind.
The Verisk CG 40-47 AI-liability pricing gap analysis published earlier in 2026 documented the industry-wide reluctance to attach class codes to AI-driven business classifications precisely because the loss-cost assumptions had no credibility. aiSure sidesteps the classification problem by operating on a bespoke-underwriting basis: every policy is individually calibrated, so there is no need to publish a class-rate table. That approach scales badly, however, and Mosaic's syndicated distribution will eventually require either a tabular pricing schema or a machine-priced quote engine. Both routes require the loss-experience credibility the book does not yet have (The Insurer, February 2026).
Regulatory Interaction With the FTC and NAIC
The regulatory environment aiSure enters is materially different from the environment that greeted early cyber. The FTC's July 7, 2026 policy statement on AI accuracy makes it a Section 5 exposure for a company to market an AI system as accurate, objective, or neutral if the system's design does not consistently deliver on that framing. That framework applies squarely to insured AI vendors whose losses aiSure is designed to cover, and it also applies, more awkwardly, to the insurer itself when Munich Re or Mosaic markets aiSure using accuracy-adjacent language.
The NAIC Model Bulletin on the Use of Artificial Intelligence Systems by Insurers, now adopted in a majority of US states, requires insurers deploying AI to maintain documentation on model validation, testing, and vendor oversight. That documentation is the same evidence base an aiSure underwriter needs to calibrate a parametric trigger for an insured. The overlap creates a peculiar dual-role structure where Munich Re, deploying its own AI systems inside its underwriting, is regulated as an AI-using insurer under the NAIC framework, while simultaneously insuring the AI errors of other companies subject to the same framework. Both roles produce discoverable model-governance documentation that could surface in a downstream enforcement action (Munich Re, July 2026).
The Allianz-Anthropic audit-ready framework that emerged earlier in 2026 tried to solve the compliance-documentation side of this dual-role problem by pre-committing to a standardized audit trail insurers could hand to any regulator on request. aiSure's underwriting file, calibrated per insured, produces something structurally similar: a per-policyholder record of the model-validation exercise that supports the parametric trigger. That documentation makes the underwriter's job at bind harder but the compliance defense at any subsequent regulatory inquiry stronger. Insurers pricing excess layers above aiSure will inherit the primary's calibration file and its regulatory posture, which is an unusual dependency for a follow market to accept (Reinsurance News, July 2026).
Mosaic's Distribution Role and MGA Economics
Mosaic Insurance's role in the aiSure program is not incidental. The Bermuda-domiciled specialty carrier operates on an MGA-plus-syndicate model, writing primary business on paper it originates and then ceding a large share of the risk into a Lloyd's syndicate structure that Munich Re participates in. The February 27, 2026 launch of the MGA program had already established the distribution pipes; the July 8 Jurecka announcement was the formal Munich Re-parent-level ratification of the program at reinsurer scale (The Insurer, February 2026).
The MGA economics inside the deal produce two distinct actuarial exposures. Mosaic collects a fronting-plus-commission fee for the underwriting work and the distribution reach; Munich Re carries the retained loss risk on the reinsured portion. For Mosaic, the risk profile is largely fee-income with a small underwriting-participation strip. For Munich Re, the risk profile is genuine parametric-loss exposure calibrated to the trigger design Berger's Insure AI team constructs. The Corgi AI-liability insurance MGA valuation environment that produced a $1.3B valuation for a startup MGA in early 2026 suggests that the market's willingness to reward AI-liability distribution capacity is already priced into equity-market comparables, which is likely part of why Mosaic moved to lock in the Munich Re relationship on an exclusive parametric-cover basis.
Munich Re's Broader AI Posture in H1 2026
The aiSure announcement fits inside a broader Munich Re posture on artificial intelligence that Jurecka has cultivated throughout the first half of 2026. Selective margin discipline, capital deployment into specialty niches where the reinsurer holds a technical edge, and an explicit orientation toward emerging risk classes have been the recurring themes in Munich Re's investor communications. The Ad-Hoc News summary of the July briefing framed the aiSure disclosure alongside Jurecka's broader statement that Munich Re would continue to prioritize technical margin over top-line growth, treating AI as a specialty class rather than a mass-market opportunity (Ad-Hoc News, July 2026).
The reinsurer's Q2 2026 combined-ratio discipline has held, and the aiSure product line sits inside a book that Munich Re can comfortably absorb without moving the underwriting-result needle in the near term. The strategic value of the product is not the near-term premium; it is the underwriting-data flywheel the parametric structure produces. Every policy Munich Re binds under aiSure adds a per-insured model-validation record to its underwriting file, and every trigger event adds a per-insured claim record to its loss file. Both feed the pricing sophistication that Munich Re will need to defend margin as the AI-liability market matures, and both put the reinsurer several years ahead of any competitor still relying on subjective proof-of-loss claims data (Munich Re, July 2026).
What Reserving Actuaries Should Watch
The reserving profile of a parametric AI-error book is unusual enough that traditional reserve-review muscle memory will mislead. The incurred-but-not-reported profile of aiSure business will look flat compared to a tech E&O portfolio, because the parametric trigger either fires and reports in a short window or does not fire at all. Case reserves will be functionally binary. Ultimate loss ratios will crystallize far faster than an actuary trained on long-tail casualty is accustomed to. All of that produces a cleaner reserve position, but it also removes the traditional cushion of IBNR that lets actuaries absorb pricing errors quietly over time.
The corollary is that a mispriced aiSure vintage will surface as an underwriting-result problem within months, not years. Triggers calibrated too tightly will over-fire and produce a loss ratio spike inside the same accident year the policies were bound. Triggers calibrated too loosely will under-fire and produce a loss ratio that looks unrealistically strong, which the underwriting team will then quietly reprice without any obvious reserve implication. Both patterns are visible in parametric weather and increasingly in parametric cyber, but neither is familiar to the specialty-casualty reserving cadre most reinsurers rely on. The cyber-AI liability digital-risk convergence pattern produced similar reserving surprises during 2024 and 2025 as parametric-cyber layers matured.
Regulatory reserving disclosure will also test the statutory framework. A parametric trigger that pays out on a benchmark breach is functionally derivative-adjacent, and the characterization of the payout obligation (loss reserve, contingent liability, or structured settlement) will vary by jurisdiction. Munich Re's German supervisory regime will treat the exposure differently than Mosaic's Bermuda framework, and the ceded flows between the two will need matching characterizations. Reserving actuaries should expect to defend the characterization in the first statutory examination cycle following launch (Munich Re, July 2026).
| Item | Detail | Source |
|---|---|---|
| Announcement date | July 8, 2026 | Ad-Hoc News, Munich Re |
| Original MGA launch | February 27, 2026 | The Insurer, Royal Gazette |
| Per-claim limit | €15M (approx $15M) | Munich Re product page |
| Named peril categories | LLM hallucination, algorithmic discrimination, copyright infringement, regulatory fines | Munich Re, Mosaic |
| Trigger mechanism | Parametric, fixed model-accuracy benchmark thresholds | Munich Re, Reinsurance News |
| Insurer partners | Munich Re (reinsurer), Mosaic Insurance (MGA distribution) | Mosaic press release |
| Munich Re unit | Insure AI, led by Michael Berger | Munich Re |
Further Reading
- FTC's AI Accuracy Policy Statement Pulls Insurer Models Under Section 5: The federal deception framework that now applies to both aiSure's insured AI vendors and to the insurer's own AI marketing language.
- AI Liability Coverage: Thin-Data Pricing Methodology for the Actuary: The core pricing problem parametric triggers try to solve by producing observable claim-frequency signals earlier than traditional proof-of-loss adjudication.
- Corgi AI-Liability Insurance and the $1.3B MGA Valuation: The distribution-side comparable that helps explain why Mosaic moved to lock in exclusive parametric-cover access with Munich Re.
- Verisk CG 40-47 and the AI-Liability Pricing Gap: The class-code and loss-cost credibility gap aiSure sidesteps with bespoke per-policy underwriting.
Sources
- Ad-Hoc News: Munich Re Plays a Selective Hand, Margin Discipline Meets AI Insurance (July 2026)
- Munich Re: Insure AI (aiSure Product Page) (July 2026)
- Mosaic Insurance: Mosaic Partners With Munich Re's aiSure to Provide Pioneering Coverage for AI Vendors (February 2026)
- Reinsurance News: Mosaic and Munich Re Introduce AI-Specific Insurance for Developers (July 2026)
- The Insurer: Mosaic Partners With Munich Re's aiSure to Launch AI Performance Cover (February 2026)
- Royal Gazette: Insurer Rolls Out Cover for AI Hallucinations (February 2026)