AIG Deploys LLM Agents at Lloyd’s via Palantir Foundry: Syndicate 2479 and the $4 Billion E&S Bet

AIG, Blackstone and wholesale broker Amwins announced the formation of Lloyd's Syndicate 2479 in December 2025, and the vehicle began underwriting on January 1, 2026 with $300 million of initial premium. What separates it from an ordinary syndicate launch is the layer underneath: risk selection runs on Palantir's Foundry platform, with large language model agents retrieving data and testing how Amwins' program portfolio sits against the syndicate's appetite.

That makes 2479 the clearest live case of generative AI sitting inside a binding decision rather than beside it. The actuarial question is what the selection does to the book it produces.

Key Takeaways

• $300 million of initial premium, expected to reach roughly $400 million by year-end, drawn from a cross-section of Amwins' approximately $6 billion in delegated authority premiums.
• AIG built an ontology giving the LLM agents access to over four million industry data points for portfolio analysis and individual risk evaluation.
• 370,000+ E&S submissions processed at Lexington at year-end 2025, up 26% year over year, which puts the 2030 target of 500,000 at 74% achieved four years early.
• Underwriting review per submission fell from three to four weeks to less than one day, alongside a 35% improvement in submit-to-bind ratios at Lexington Middle Market Property.
• The $4 billion new E&S premium target for 2030 is a premium goal, not a result: no accident year on the AI-selected book has developed yet.

What Syndicate 2479 Actually Is

The vehicle uses Lloyd's London Bridge framework, which lets third-party capital participate in Lloyd's market risk through a protected cell. Funds managed by Blackstone provide that capital alongside Amwins and AIG. Talbot Underwriting, AIG's existing Lloyd's managing agent, runs the syndicate; it already operates Syndicate 2001.

Distribution runs through Amwins, which channels a diversified slice of its delegated authority book into the syndicate. That book is roughly $6 billion in premiums written by coverholders with binding authority, so 2479 sits on top of a large and varied pool of specialty risk rather than building a submission flow from scratch.

The technology layer is what the announcement actually leads with. Foundry is not a document summarizer. It is an ontology platform: submissions, loss history, exposure data, policy forms and industry databases get organized into a structured map where relationships between objects are explicitly defined. AIG built such a map of the Amwins portfolio, connecting programs to risk characteristics and individual submissions to historical loss patterns.

AIG CEO Peter Zaffino described the syndicate as representing "the next level of innovation, technical modeling" in portfolio underwriting. The four million data points are not a static reference file; they feed the ontology continuously, so an agent evaluating one program can place it against market-wide benchmarks and portfolio aggregation limits at the same time.

Period	Milestone
Late 2024	AIG Assist debuts inside North America Financial Lines for private and non-profit business
Q1 2025	Early results described as “very promising” by CEO Zaffino; 100% of applicable Financial Lines submissions processed by AI
Q3 2025	Middle market property and casualty rollout begins at Lexington Insurance Company
September 2025	AIG Investor Day features Anthropic CEO Dario Amodei and Palantir CEO Alex Karp alongside Zaffino; 500,000-submission target by 2030 announced
End of 2025	Full Lexington rollout complete; 370,000+ submissions processed (26% YoY increase); Syndicate 2479 announced
January 2026	Syndicate 2479 begins underwriting $300M portfolio
February 2026	Q4 2025 earnings call: Zaffino says outcomes are “beyond expectations”; acceleration timeline pulled forward
March 2026	McGill and Partners collaboration announced: $1.6B specialty portfolio with agentic AI and Palantir Foundry
2026 (planned)	Full rollout across North America, UK, and EMEA commercial lines; Claims by AIG Assist scaling

Throughput Is the Mechanism, and Throughput Moves the Mix

The syndicate is the visible end of a platform AIG has been scaling internally since late 2024, and the platform's disclosed metrics are all throughput metrics. Lexington processed more than 370,000 E&S submissions in 2025, up 26% year over year. Review time per submission fell from three to four weeks to less than a day, according to Carrier Management reporting. Data accuracy moved above 90%, from roughly 75% before deployment.

The number that matters actuarially is the 35% improvement in submit-to-bind ratios at Lexington Middle Market Property. Faster review is an expense story. A submit-to-bind shift of that size is a statement about which risks now get bound, which is a mix story.

That distinction decides how much of the pre-2025 history is usable. Loss development factors fitted to a book selected by underwriters working three-week queues describe a portfolio assembled under a different filter. If AI triage is genuinely screening out weaker submissions, the AI-selected book should develop differently, and the 370,000-submission portfolio cannot be treated as a scaled-up version of the old one for reserving purposes.

The contrast with Chubb sharpens the point. Chubb has tied its AI deployment to a 20% headcount reduction target and 85% process automation; AIG is writing more business with the same team, reviewing 100% of private and non-profit business submissions in Financial Lines without adding underwriters. Both compress the expense ratio. Only AIG's version also changes the composition of the risk being reserved.

The Orchestration Layer Is the Part That Does Not Exist Yet

AIG has said it is building an orchestration layer to coordinate multiple agents across the enterprise, and has not yet fully deployed it. Until it exists, the architecture is a set of strong single-workflow deployments rather than the closed loop the ontology concept implies. Claims by AIG Assist is scheduled to scale during 2026; while claims and underwriting sit on separate rails, the loss feedback that would validate the selection thesis does not reach the pricing side quickly.

The extension into the subscription market adds exposure before that validation arrives. Under the March 2026 collaboration with McGill and Partners, AIG will deploy 25% follow capacity across up to $1.6 billion of McGill's specialty gross premiums written. Follow underwriting has historically been decided account by account. Pre-securing a quarter of the line across a whole portfolio moves the judgment upstream into the ontology's representation of that portfolio, and any error in how aggregation is modelled is now committed capacity rather than a declined risk.

The binding constraint is that none of this has produced a loss year. Syndicate 2479 began underwriting on January 1, 2026, so its first accident year is still open. The $4 billion 2030 target measures premium through the pipeline, not the result of what the pipeline selected. Whether the AI-chosen book develops better than the book it replaced is the only test that settles the thesis, and it is several years of development away.

Sources