Authorization of external customer and broker agents
Insurers must distinguish a trusted agent acting for an insured from a bot without authority; access should be scoped by task and identity.
Executive question
Who delegated which authority to which agent for this exact insurance action?
What to measure
- Authenticated mandate coverage
- Failed and revoked delegation rate
- Proof of authority for cancellations/claims
- Cross-entity trust incidents
Relevant products and standards
A working draft, stewarded by Marrow, for how AI agents quote, compare and buy UK regulated insurance, with rules on consent, disclosures, data use and audit.
Model Context Protocol: the standard way an AI agent connects to tools and data. Created by Anthropic (Nov 2024); donated to the Linux Foundation's Agentic AI Foundation on 9 Dec 2025.
Personal Agent Protocol: a draft (v0.1) for a customer's personal agent to identify itself to a business, get consent and open a session. Announced by Sierra and Meta on 6 Oct 2026 with six named partners; the 9 Oct draft lists seven initial partners and 35 design partners.
Relevant examples, not documented integrations, completeness, certification or endorsement. Listed alphabetically.
Evidence and current limits
Strategic pathway; not an observed market-wide adoption rate.
Draft analysis, under editorial review.