Meta and Sierra announce Poppy, a protocol for personal AI agents

Meta and Sierra have announced Poppy, a protocol intended to let personal AI agents identify themselves, obtain customer permission and complete tasks across business websites and APIs. The practical question is whether this can make agent-led transactions easier to authorise without making account access harder to control.

Watch Desk analysis

What happened

A report by Nekuda says Poppy builds on OAuth and is intended to help agents operate across business websites, APIs and other agents. It names OpenAI, Visa and Stripe among dozens of partners alongside Meta and Sierra.

The report says companies implementing Poppy will need to coordinate their CDN, endpoint and page-security controls so authorised agents can carry out permitted actions. It frames the protocol as a response to security, authorisation and liability challenges as consumer agents act on people’s accounts.

Why it matters

An agent that can take action for someone needs more than a clever prompt. Services need a way to recognise the agent, establish what the customer has allowed and limit what it can do. A shared protocol could give businesses and agent makers common ground, rather than leaving every integration to improvise its own permissions.

That is the promise, not proof of adoption or safety. The report gives no implementation examples or independent assessment of how Poppy handles misuse, disputes or revoked permission. A long partner list is encouraging; the security details will have to do the heavier lifting.

Our read

Poppy is aimed at a real bottleneck for useful AI agents: getting permission to act across services, not merely generating another answer. The interesting test is whether it gives customers clear, enforceable control, rather than making it easier for businesses to wave an agent through.

What to watch

• Which partners implement Poppy, and what customers can do with it.
• How permission is granted, limited and withdrawn.
• How the protocol handles unauthorised actions and disputes.

Discussion spark: Should personal AI agents use one shared permission standard across services, or should each business set its own rules even if that makes agents less useful?

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.