Sierra and Meta unveiled a shared way for customers' AI agents to prove who they are to businesses. Here is what owners should do before more of them show up.
Sierra and Meta announced the Personal Agent Protocol, or PAP, on October 6, 2026. It is an open standard, meaning a public set of rules anyone can use. It covers how a customer's AI agent proves who it is to a business and how it acts for that customer. Walmart, Shopify, Stripe, Rocket, Genesys and Instinct are the founding partners.
Sierra says control is split in two. Customers decide what access their agent gets. Businesses decide which actions that agent may take. For a business, that could make it easier to tell an agent working for a real customer from an unidentified bot.
The problem is already real. Meta launched its Muse shopping agent on September 8. On September 20, Amazon blocked it. Amazon said unauthorized agent access breaks its terms and that the agent does not identify itself as an agent. Other companies reportedly blocked Muse for the same reason.
PAP is early and is not an industry-wide standard. Sierra and Meta plan to publish the first draft, called v0.1, later this month. Design workshops and a reference implementation, a working example for builders to follow, are also planned. Payments are not in the first version. Sierra lists them as future work, so identity and permissions come first and money comes later.
The field is also crowded. Walmart and Shopify take part in the Universal Commerce Protocol with Google, Target, Etsy and Wayfair. Stripe co-developed the Agentic Commerce Protocol with OpenAI. Visa's Trusted Agent Protocol covers overlapping ground from the payments side.
No one needs to adopt anything yet. The practical step is preparation:
Three things will show whether PAP catches on. First, whether the v0.1 specification appears this month. Second, whether the partner list grows beyond the founding group. Third, whether Amazon and other retailers that blocked Muse engage with the standard. Until those answers arrive, expect more than one set of rules to compete.
More customers will send AI agents your way, so decide now what an agent may do and ask your software vendors about their plans.
Visa tested 22 AI models on realistic shopping tasks. They passed just over half, and often bought the wrong thing anyway.
Read more: Visa tested 22 AI shoppers, and most still make costly mistakesAI researcher Andrej Karpathy built a video site in one weekend with Claude. He says it beats YouTube's own picks, and it hints at how fast small software projects can move.
Read more: Karpathy built a video site in one weekend with Claude's helpDARPA wants AI to hunt for design flaws in stock, bond, energy and cloud markets, so enemies cannot use them to cause economic damage that reaches households.
Read more: Pentagon wants AI to find weak spots in stock and energy markets
