AXA has moved its Global AI Hub into production across five entities, with AXA XL among those building claims and email applications on the shared infrastructure. The rollout marks a shift from isolated AI pilots to an industrialised, governed deployment model at a time when nearly 80% of large insurers have adopted AI-assisted processing workflows, according to Datos Insights research.
What the hub actually does
The hub provides standardised infrastructure for AI agents, with governance, compliance and human oversight built into the architecture rather than layered on afterwards. AXA entities in Germany, France, Switzerland and the United Kingdom are developing applications including motor claims automation, customer email processing and enterprise knowledge management. The design lets individual entities build on shared foundations instead of each constructing the same underlying infrastructure independently.
The platform is vendor-agnostic and supports multiple large language models across different use cases. That approach reduces dependency on any single AI provider and allows AXA to switch or combine models as the technology evolves. Publicis Sapient provides the engineering and AI platform expertise to accelerate the build. The first platform version went live in July.
Governance as operational architecture
AXA has embedded what it calls FinOps for cost management, SafetyOps for AI safety monitoring, and compliance oversight throughout the AI lifecycle. Matthieu Caillat, group chief technology and AI officer at AXA, said the priority was combining safe and efficient AI with governance and practical business applications. "Publicis Sapient brings together specialist AI expertise, enterprise AI solutions and global experience that will help us accelerate AI adoption consistently across our organisation, while maintaining the governance and flexibility required in a highly regulated business," he said.
The FCA has not introduced AI-specific rules, but the Consumer Duty and the Senior Managers and Certification Regime already apply to AI-assisted decisions. The Treasury Select Committee's January 2026 report noted that more than three-quarters of UK financial services firms use AI for core functions and called on the FCA to clarify how accountability under the SMCR operates where AI systems perform functions traditionally subject to direct human oversight. AXA's human oversight component addresses that question at the infrastructure level, though how it functions in practice is not yet publicly detailed.
How firms demonstrate accountability for AI-assisted decisions to regulators and clients is becoming a practical differentiator. Embedding governance in the platform architecture rather than retrofitting controls later is a more defensible model - but its effectiveness will depend on how the human oversight points are designed at the application level, not just the infrastructure level.
Scale and what follows
The Publicis Sapient partnership builds on several years of AI for Insurance investment at AXA. The group renewed a five-year AI partnership with Shift Technology in March, covering fraud detection and claims automation across 15 countries. In July, it deployed Microsoft 365 Copilot to its global workforce. The Global AI Hub sits above those individual tools as the infrastructure layer through which future AI agents will be deployed and governed.
Aviva has deployed AI summarisation in life underwriting, and Legal and General extended Microsoft Copilot to its entire workforce in June. The direction across the largest carriers is consistent: AI Agents & Automation are moving from pilot to production, and the governance question is shifting from whether to build controls to how those controls function in live underwriting and claims environments.
Why this matters for brokers
Brokers placing commercial or specialty risk with AXA XL are dealing with claims workflows and servicing models that are changing structurally as these systems move from pilot to production. That is not a reason for concern in itself - AI-assisted claims processing, where well-governed, can accelerate turnaround and reduce handling errors. But it is a reason to ask specific questions.
For brokers placing business with AXA XL, asking how the hub's claims automation tools interact with human claims handlers - and at what point in a claim's lifecycle a human decision is required rather than an AI-generated recommendation - is a reasonable and increasingly standard due diligence question. The SMCR requires a named individual to be accountable for material decisions. Understanding how that accountability is preserved when AI is embedded in the claims workflow is both a regulatory question and a practical one for brokers whose clients expect to know who is responsible when a claim is handled.
Your membership also unlocks:
