Marrow
The standard

A2A protocol

The A2A (Agent-to-Agent) Protocol is a set of open standards defining the canonical schema, behavioural rules, and authority model for AI agent-mediated insurance transactions in the UK. It specifies a canonical risk schema per line of business, a Core Patterns standard governing agent behaviour, and a Consent and Authority standard governing what agents may do and how they prove it.


What is a2a protocol?

The A2A protocol is organised as a stack of composable standards. At the base is a governance layer covering versioning, conformance rules, and alignment to external standards including Polaris UK for code lists, Defaqto for product taxonomy, FCA rules for conduct, and UK GDPR for data handling.

Above the governance layer are three planes that apply across all lines of business. Plane 1, Interactions, defines the lifecycle of interactions an agent can have: Inform, New Business, Mid-Term Adjustment, Renewal, Claims, Cancellation, and Servicing. Plane 2, Information, classifies the types of data an agent can read by access authority: Public, Policyholder-authenticated, Consented enrichment, Derived, and Regulatory disclosure. Plane 3, Core Patterns, defines the cross-cutting behavioural rules every conformant agent must follow: grounding and no-fabrication, assume-infer-confirm, the advice-versus-information boundary, disclose-and-acknowledge, honest absence, data protection, vulnerability and Consumer Duty, handover, and audit.

Line-of-business standards sit on top of these three planes. The Home standard (buildings and contents) is live in production with a Tier-1 UK carrier at version 1.0. Motor is at hardened draft v0.2. SMB commercial lines, pet, and travel are at draft v0.1.

The protocol is published as an open standard; Marrow stewards the drafting but does not assert proprietary control over the interface. Carriers, AI platforms, and regulators are invited to participate in its development. The protocol is also referred to informally as the Marrow Protocol, reflecting Marrow's role as steward.


Why does a2a protocol matter for insurance?

Without a shared protocol, every AI-to-insurer connection requires bespoke integration: each carrier behaves differently, each agent re-implements compliance independently, and a regulator has no single document to read to understand how a transaction was governed.

The A2A protocol is designed to solve that. A carrier that maps its products to the A2A canonical schema, and an AI agent that consumes it, interoperate without custom code. The open standard also provides the artefact that AI platforms and the FCA need to see to trust that agentic insurance transactions are governed consistently.


Related terms

AI compliance layer

The regulated Rail that enforces the A2A protocol in real time.

Insurer MCP endpoint

How a carrier's systems are exposed to AI agents under the protocol.

Quote-to-bind

The full New Business transaction flow defined by the A2A protocol.

Step-up confirmation

The Tier A authority mechanism defined in the A2A Consent and Authority standard.

Conversational underwriting

The risk-data assembly approach the protocol defines for New Business.

Hallucination firewall

The enforcement mechanism for the protocol's no-fabrication rule.


Source

The Standard (Marrow, 2026)

The Standard is the primary reference for all A2A terms. Marrow stewards the current draft.

Last updated 2026-06-18

All terms