Artificial Intelligence at AIG

Headquartered in New York, AIG is a worldwide insurer offering business and private property-casualty protection throughout Liability, Financial Lines, Property, Global Specialty, Crop Risk Services, Personal Lines, and Accident & Health. As of December 31, 2025, the corporate employed roughly 22,100 individuals throughout 45 nations and underwrites enterprise in additional than 200 nations and jurisdictions via its world community.​

AIG reported $3.1 billion in internet earnings for full-year 2025. General Insurance recorded $23.7 billion in internet premiums written, whereas its mixed ratio improved by 1.7 proportion factors yr over yr, from 91.8% to 90.1%. Global Commercial internet premiums written increased 4% on a reported foundation to $17.4 billion, supported by 9% progress in new enterprise.

AIG has recognized AI deployment throughout underwriting and claims as a key strategic focus, stating in its 2025 Annual Report that it’s deploying and scaling agentic AI options to hurry processes and enhance decision-making in these areas. In 2025, the corporate considerably advanced this technique via expanded partnerships with Palantir, Anthropic, AWS, and Google, embedding AI capabilities throughout underwriting and claims.

This article examines two use circumstances illustrating how AIG is making use of this funding inside its operations:

• Accelerating Underwriting Submission Triage via AIG Assist — AIG makes use of massive language fashions to learn, prioritize, and summarize incoming submissions, enabling underwriters to guage extra alternatives with out compromising underwriting self-discipline.
• Speeding Capacity Deployment via Portfolio Ontologies with Palantir — AIG makes use of machine-readable representations of its insurance coverage portfolios to allow AI brokers to guage danger and assist quicker capital allocation selections, with potential purposes throughout its operations and third-party capital and distribution partnerships.

We start with AIG Assist and the appliance of generative AI to underwriting submission triage.

Accelerating Underwriting Submission Triage via AIG Assist

Complex business underwriting creates a bottleneck that extra headcount alone can’t eradicate. AIG’s investor supplies indicate that manually reviewing a posh business submission can take three to 4 weeks. They additionally reveal that Lexington, AIG’s extra and surplus traces enterprise, certain insurance policies for under about 2% of the roughly 300,000 new-business submissions it obtained in 2024. Hiring extra underwriters might improve assessment capability, but it surely doesn’t basically scale back the time required to evaluate every submission or determine the alternatives most probably to warrant protection.

Video: AIG Underwriter Assistance in Action | AIG at AIPCon 7 (Source: Palantir)

A business insurance coverage submission sometimes arrives as an unstructured assortment of dealer cowl letters, statements of worth, loss runs, and supplemental purposes, all in inconsistent codecs. Underwriters have historically needed to learn and reconcile these paperwork manually earlier than deciding whether or not — and on what phrases — to cite the danger. AIG’s investor supplies illustrate the ensuing funnel at Lexington, its extra and surplus traces enterprise: roughly 300,000 new-business submissions in 2024 resulted in about 6,700 certain insurance policies, equal to a bind price of roughly 2% and roughly $1 billion in new-business premium.

AIG’s 2030 ambition for Lexington — 500,000 submissions, a 6% bind price, and $4 billion in new-business premium — requires the corporate each to course of extra submissions and convert a better proportion of them. Expanding the underwriting workforce may improve assessment capability, however it could not basically change the time required to evaluate every submission. AIG developed AIG Assist to deal with that constraint.

Screenshot: AIG’s illustration of how AIG Assist synthesizes, prioritizes, and prepares submission information for underwriter assessment (Source: AIG Investor Day 2025 presentation, March 31, 2025, PDF p. 117)

​AIG Assist features a patent-pending capability known as Auto Extract, which makes use of massive language fashions to extract structured information from unstructured submission paperwork throughout diversified codecs. The device then assesses incoming submissions in opposition to the related enterprise’s said danger urge for food, prioritizes them for assessment, and produces curated summaries for underwriters. This modifications the preliminary triage course of from reviewing submissions within the order obtained to specializing in alternatives that seem to suit the insurer’s underwriting standards.​

In 2026, AIG described the following section of the know-how being developed with Palantir and Anthropic, as a multi-agent underwriting system. On AIG’s Q1 2026 earnings name, CEO Peter Zaffino outlined a proposed structure during which purpose-built brokers may carry out submission ingestion and information extraction, consider dangers in opposition to underwriting pointers, benchmark pricing in opposition to portfolio targets, and synthesize their findings for an underwriter.​

AIG can be developing an orchestration layer to coordinate these brokers and their handoffs. The firm expects the system to complement underwriters’ evaluation whereas sustaining human oversight of danger evaluation, pricing, and protection selections. AIG characterised this multi-agent structure as a system below improvement slightly than a totally deployed manufacturing functionality.​

AIG Assist’s deployment has expanded in levels slightly than launching all over the place at as soon as:​

• AIG Assist entered manufacturing in Financial Lines, together with its Private Not-for-Profit enterprise, the place AIG later reported that the system was reviewing 100% of relevant submissions.
• AIG subsequently started deploying the device throughout Lexington’s middle-market property and casualty operations. The firm mentioned in November 2025 that it anticipated to complete deployment throughout Lexington’s remaining wholesale enterprise by year-end.
• AIG additionally accelerated its deliberate deployment throughout North American, U.Okay., and EMEA business traces by six months.
• On the claims facet, AIG has piloted related document-extraction capabilities supposed to shorten the interval between receiving a primary discover of loss and issuing a protection letter.

For underwriters, the operational change is a shift from a assessment queue constrained by the point required to learn and reconcile paperwork to 1 during which extraction, summarization, and an preliminary evaluation can happen earlier than human assessment. AIG describes the ensuing change to an underwriter’s day-to-day work alongside a number of traces:​

• Instead of starting with an unsorted assortment of paperwork, an underwriter receives a prioritized submission with extracted information and a machine-generated abstract.
• Underwriting time can shift away from information entry and doc reconciliation and towards danger choice, pricing, protection, and policy-structure selections.
• The system is designed to assist slightly than exchange underwriting judgment, with underwriters remaining liable for consequential danger and protection selections.
• On the claims facet, an analogous method may speed up the preliminary doc assessment between the reporting of a loss and the issuance of a protection letter.

AIG has supported its broader digital transformation with substantial funding. The firm reported in 2024 that it had invested roughly $300 million in information, digital workflows, AI, and expertise over the previous two years, following greater than $1 billion in foundational information know-how funding over 5 years. These figures cowl AIG’s wider know-how program slightly than AIG Assist alone.

​In an early AIG Assist deployment, AIG’s investor supplies state that submission turnaround fell from three to 4 weeks to lower than someday. The proportion of relevant submissions reviewed elevated to 100%, slightly than a filtered subset. Zaffino additionally reported that data-extraction accuracy improved from roughly 75% to greater than 90%, alongside a considerable discount in processing time.

​AIG continued to report progress via subsequent quarters. By year-end 2025, Lexington had obtained greater than 370,000 submissions — up 26% yr over yr and representing substantial progress towards, however not completion of, its 2030 ambition of 500,000. AIG additionally reported a 35% enchancment within the submit-to-bind ratio for Lexington’s middle-market property enterprise following the rollout of AIG Assist.​

On AIG’s Q1 2026 earnings name, Zaffino said the device had helped Lexington’s middle-market property enterprise quote 30% extra submissions, scale back underwriters’ time to cite by 55%, and improve the variety of submissions certain by roughly 40%. These figures recommend that the system is affecting each processing capability and business conversion. However, AIG has not disclosed the underlying volumes, measurement intervals, or contribution of different operational modifications in sufficient element to isolate the impact of AIG Assist.​

AIG reported these efficiency figures in earnings calls and investor supplies, and so they haven’t been independently validated. The 2030 submission, bind-rate, and premium figures additionally stay ambitions slightly than achieved outcomes. Nevertheless, the disclosures present a transparent deployment sample: AIG launched the know-how in chosen underwriting companies, prolonged it throughout extra traces, and reported enhancements in assessment pace, submission protection, and conversion as deployment progressed.

Speeding Capacity Deployment via Portfolio Ontologies with Palantir

AIG’s second AI use case addresses a broader downside than accelerating the assessment of particular person submissions: making a structured, queryable illustration of a complete portfolio in order that AI methods can consider exposures and assist underwriting and capacity-deployment selections.

​AIG refers to this illustration as an ontology. Built on Palantir’s Foundry platform, it integrates info corresponding to insured dangers, exposures, coverage limits, attachment factors, modeled losses, and underwriting guidelines. Rather than creating a separate account-level integration for each portfolio, AIG has utilized the identical ontology-building method throughout a number of more and more complicated business preparations.

​

Screenshot: Editorial illustration of AIG’s reported ontology-based capacity-deployment workflow; not a picture of AIG’s manufacturing interface. Sources: AIG and Palantir public bulletins.

AIG has disclosed three purposes of its ontology-based method:

Everest renewal-rights transaction

• In October 2025, AIG agreed to accumulate renewal rights for many of Everest Group’s world retail business insurance coverage portfolios, representing roughly $2 billion in premium.
• Everest retained the liabilities and claims-administration duties related to its current insurance policies, whereas AIG gained the chance to supply protection to eligible accounts at renewal.
• To assist the transaction, AIG developed an “Everest ontology” — a digital mannequin of the portfolio that enabled its underwriters to guage account limits, attachment factors, and pricing and decide how the acquired enterprise may match inside AIG’s current portfolio.

Lloyd’s Syndicate 2479

• AIG subsequently labored with Palantir, Amwins, and funds managed by Blackstone to ascertain Syndicate 2479, a special-purpose automobile at Lloyd’s managed by Talbot Underwriting.
• The syndicate was established to underwrite $300 million in premium from a diversified portion of Amwins’ roughly $6 billion delegated-authority portfolio starting January 1, 2026.
• AIG used Palantir Foundry to validate its portfolio evaluation earlier than launch and developed an ontology that enabled massive language fashions to entry greater than 4 million business information factors.
• Importantly, AIG described the portfolio evaluation as already accomplished, whereas positioning using a number of brokers to retrieve information, consider danger traits, and check packages in opposition to the syndicate’s danger urge for food as a longer-term functionality.

McGill and Partners collaboration

• In March 2026, AIG announced a collaboration with specialty dealer McGill and Partners below which AIG expects to offer 25% capability throughout as much as $1.6 billion of McGill’s specialty gross premiums written.
• After analyzing the portfolio, AIG developed underwriting standards to assist real-time underwriting by way of McGill’s digital broking platform.
• AIG and Palantir additionally constructed an ontology supposed to offer near-real-time info on exposures, deployed limits, modeled danger, and losses, enabling AIG to observe portfolio efficiency and handle its capability on an ongoing foundation.

For portfolio and capability managers, the supposed workflow change is much like the change AIG Assist brings to submission triage, however operates at the extent of a complete guide of enterprise. AIG and its companions describe the ensuing capabilities as follows:

• Exposure, restrict, modeled danger, and loss information could be built-in right into a shared portfolio mannequin slightly than assembled manually from a number of methods for every assessment.
• Accounts and packages could be evaluated persistently in opposition to outlined underwriting standards and portfolio-level danger urge for food.
• Managers can monitor the impact of deploying extra capability as portfolio information modifications.
• AI brokers may retrieve and consider related info throughout the portfolio, a functionality AIG has characterized as a longer-term aim slightly than a present manufacturing functionality.

The sources don’t set up that these capabilities eradicate account-level assessment or autonomously allocate capital. Instead, AIG presents them as decision-support infrastructure supposed to make portfolio evaluation quicker, extra constant, and extra attentive to altering publicity information.​

The three preparations additionally use completely different insurance coverage and capital buildings. In the Everest transaction, AIG obtained renewal rights and mentioned it may write qualifying insurance policies on its current stability sheet with out requiring incremental capital. In Syndicate 2479, Amwins and funds managed by Blackstone present third-party capital for a portfolio managed by AIG via a Lloyd’s automobile. In the McGill collaboration, AIG supplies capability to qualifying dangers distributed via the dealer’s platform.

​Palantir provides the Foundry know-how for organizing and analyzing portfolio information. AIG contributes its underwriting standards, portfolio evaluation, and danger urge for food. Amwins and McGill present entry to distribution and portfolio information, whereas Amwins and Blackstone additionally present capital to Syndicate 2479. Talbot serves as Lloyd’s managing agent for that syndicate.

​Zaffino has described the target as combining AI-enabled portfolio insights with underwriting self-discipline to deploy capability extra rapidly. Palantir CEO Alex Karp has equally positioned the Foundry deployment as a approach to assist new partnership buildings and working efficiencies.

​This use case is at an earlier stage than AIG Assist and has much less disclosed efficiency information. Syndicate 2479 had operated for under a brief interval when AIG first mentioned its progress publicly, whereas the McGill collaboration was introduced in March 2026. AIG and its companions haven’t but disclosed loss ratios, retention charges, capability utilization, or different underwriting-performance measures attributable to the ontology-based course of.​

The out there proof, due to this fact, establishes the construction and supposed working mannequin, slightly than the long-term monetary outcomes. AIG has demonstrated that it will probably apply a repeatable ontology-building methodology throughout a number of portfolios: first, to investigate renewal alternatives acquired from one other insurer; then, to assist a third-party capital automobile; and eventually, to handle capability via a digital brokerage relationship. Whether that method improves underwriting profitability or portfolio efficiency at scale stays to be demonstrated.

​