Meta and Sierra Propose OAuth-Based Sign-In Standard for AI Agents
The Development
Meta Platforms and Sierra have proposed the Personal Agent Protocol, a sign-in standard for AI agents acting on behalf of users. Built on OAuth, it defines three session tiers: guest, read-only, and write. Each grants an agent progressively greater access to a user's account at a service. The proposal also lays groundwork for scoped tokens, which restrict what an agent can do within a session, and for agent-specific logging, so services can distinguish and audit agent activity separately from human activity. Sierra builds conversational AI agents for customer service, giving the proposal a direct line to brand-owned customer interactions. The proposal is a draft standard, not a ratified one. Neither company has announced adoption timelines, and no other major platform has publicly endorsed it.
Our Take
Agent authentication is the unglamorous layer that decides who captures agent-mediated demand. Today, consumer agents either browse as anonymous guests or borrow credentials in ways no security team should tolerate. A tiered standard turns agent access into something brands can price, restrict, and measure. Meta's involvement matters because it pairs the largest consumer identity graph with a customer-service agent vendor, so adoption has distribution before it has consensus. The risk is gatekeeping: a standard co-authored by a platform owner will reflect that platform's interests in how identity, consent, and logging work. Brands that wait for ratification will be configuring access policy on someone else's defaults.
What Changed
Agents acting for consumers can now authenticate to a brand's systems through a defined, permissioned handshake rather than scraped logins or shared credentials. Brands gain the ability to grant, limit, and audit agent access per session tier, separating agent traffic from human traffic at the identity layer.
Marketing Impact
CRM and customer-experience teams inherit a new access-policy decision: what an agent may read versus change in a loyalty or service account. Marketing operations and martech owners must plan for agent-labelled traffic in analytics, since logging at the protocol level will separate agent sessions from human behaviour in attribution and personalisation data.
Competitive Implication
Brands with clean first-party identity and well-scoped account APIs gain structural advantage, because agents can transact with them safely while competitors with fragmented logins get skipped or throttled. Vulnerable: brands whose account systems cannot issue scoped permissions, and vendors whose personalisation depends on assuming every session is a human.
Strategic Outlook
Expect competing proposals from Google DeepMind, OpenAI, and Anthropic within two quarters, followed by a negotiated convergence around OAuth extensions in 2027. Agent-session tiering will become a standard line item in customer-platform RFPs by mid-2027, and brands without agent-ready identity will see agent-mediated purchases route to competitors.
The Exploit
Action Item
CRM and martech leads should task identity engineering this quarter with mapping which account actions belong in guest, read-only, and write tiers, then shortlist the three highest-value agent use cases before Q1 2027 planning closes.