Scor's AI underwriting pilot stops at level three, with a human sending every offer, Léger says

Speaking on Monday at the Singapore International Reinsurance Conference, Scor's group chief executive Thierry Léger described four levels of AI governance that the reinsurer applies to underwriting, InsuranceAsia News reported. At level three, the system can do all of an underwriter's work and compose a quote, but a human underwriter makes the final call and sends the offer. At level four, the system decides and sends offers to clients on its own. Scor has stopped at level three and wants to learn from it, he said.

Léger expects the industry to go further and predicted that AI will in time send offers to clients autonomously. He called the move from level three to level four "a big step" that Scor wants to keep under control. Rather than putting up barriers, he argued, companies need "clear guardrails" to support AI's development.

Scor intends to register every agent that does work for it and will not let an unregistered agent operate, Léger said, adding that managing AI agents is unfamiliar territory and that leaders do not yet know their strengths and weaknesses. He also pointed to reliability limits. Given the same data five times, a model may return slightly different answers, and it may not retain how it reached a result, so a wrong answer cannot be questioned the way a person's reasoning can. He wants protocols that require agents to log their decisions.

On current use, Léger said Scor relies on AI heavily in contract management. In underwriting, the company has built a pilot for augmented underwriting that ingests data, reviews accumulations and supports some pricing decisions. He gave no outcome figures for either, and said the change will have a large effect on the workforce and will test leaders.

The remarks extend what Scor told reporters at the Monte Carlo Rendez-Vous in September, when Léger said the group was leaning fully into AI and that an AI-driven terrorism underwriting tool was close to going live. Jean-Paul Conoscente, chief executive of Scor Global P&C, said then that Scor planned to begin deploying AI in property catastrophe underwriting from 2027. Léger described governance as one of his biggest AI concerns and noted that AI-related lawsuits now number in the hundreds.