The Personal Agent Protocol (PAP) is an open standard for how a person's AI agent identifies itself to a business, gets permission, and acts. Sierra and Meta announced it on October 6, 2026. It is still a draft: the project's documentation showed Draft 0.3 on October 31, and it has a published path to 1.0. It does not cover product discovery or payment. It covers the relationship between the agent, the person, and the company.
How It Works
| Role | What it does |
|---|---|
| Personal agent | Acts for a person and always identifies itself |
| Company | Declares how agents can interact with it: sign-in options, channels, and access levels |
| Person | Signs in to give the agent limited access to their account |
Access runs along a range. An agent can work anonymously for something like checking availability. Signing in lets it see account data, and the customer decides between read-only and write access. Some actions, such as buying a ticket, can require approval. Before sign-in, the company sees only an ID that contains no personal information, stays the same over time, and differs from one company to the next.
A company can start with its website, its APIs, or its own agent, and add the others later. PAP builds on OAuth, JWTs, HTTPS, OpenAPI, and MCP instead of replacing them.
Who Is Backing It
Sierra's post names Genesys, Instinct, Rocket, Shopify, Stripe, and Walmart as partners. Lists differ by source: a Meta executive's post adds NiCE and Decagon and leaves out Instinct. Launch posts do not name OpenAI, Anthropic, Amazon, Google, Zendesk, or Salesforce. That matters because a standard for personal agents needs the companies that make them. Decagon also released its own related protocol, PACT, and says it is joining the PAP working group.
Where PAP Fits
| Standard | Question it answers |
|---|---|
| MCP | How does an agent call a tool? |
| UCP, ACP | How does an agent run a commerce flow or checkout? |
| AP2 | How is a buyer's instruction proven? |
| x402 | How does an agent pay for a request? |
| PAP | Who is this agent acting for, and what may it do at this company? |
What Companies Should Do
- Do not build for 1.0 yet. The draft number has already moved, and more changes are likely.
- Decide your policy first. List what an anonymous agent may do, what needs sign-in, and what always needs a person. This work carries over to any standard.
- Read the draft if you run customer accounts. Support, travel, and finance have the most to gain from clear agent access.
- Track adoption by personal agent makers. Support from ChatGPT or Claude would matter more than another business partner.
Methodology
Sources are Sierra's announcement, the PAP documentation, and eesel's comparison of partner lists. The table of standards is our own summary. Status as of October 31, 2026.
Following This With Presenc AI
Presenc AI is building tools to show how AI systems find, describe, and act on brands. Join the waitlist for access.