Agentic insurance

The A2A protocol defines agentic insurance as requiring four interdependent layers. First, a canonical risk schema per line of business - motor, home, pet, travel, SMB commercial - defining the structured data required for a quote. Second, an AI compliance rail, the regulated intermediation layer that sits between the agent and carrier systems, validates every field against product rules, strips PII, enforces disclosures, and writes the audit record. Third, a scoped authority model that maps every action an agent can take to a risk class, with escalating proof requirements. Fourth, an audit ledger recording every enrichment call, every enriched-versus-attested fact, every disclosure and acknowledgement, and every authority exercised.

The term 'agentic' distinguishes this from earlier automated insurance models. An agent is not a rule-based bot running a fixed script; it is an AI system capable of reasoning over incomplete information, populating a risk profile from authoritative sources using the look-up-not-ask principle, and handling exceptions through a defined handover pattern. The agent communicates with the compliance rail, never directly with the carrier or underlying data sources.

Agentic insurance is not unmonitored automation. The A2A Consent and Authority standard defines two conformance tiers. Tier A is the default: the customer provides fresh, human-present confirmation before every contract-changing or payment action. Tier B is an opt-in mode in which a customer pre-authorises an agent to act autonomously within defined, capped limits - for example, renewing a policy if the premium is below a specified ceiling. Tier B carries heavier conformance requirements and stronger agent identity obligations.