Story
October 7, 2026

Meta’s AI-agent protocol tries to put guardrails on a booming bot economy

Meta, Walmart, Stripe and other partners are backing an open protocol for personal AI agents as businesses seek protection from scraping, fraud and unaccountable automated access.

Meta’s push to make personal AI agents useful in everyday commerce is colliding with a basic problem: the bots can move faster than the rules governing them.

In early September, Meta launched Muse, a personal agent that rapidly climbed Apple’s App Store rankings and, according to Meta, reached millions of U.S. users. The company says people are using it for tasks from enrolling children in classes to buying gifts and arranging overdue dentist appointments. But its rise has also sharpened corporate anxieties over automated access.

Amazon has blocked Meta’s agents over concerns about website scraping. It also blocked Perplexity’s agent and sued the startup in November, alleging that its agents were concealed to keep scraping Amazon without permission; Perplexity called the case a “bully tactic.”

A month after Muse’s debut, Meta joined Walmart, Stripe, Sierra and other companies to propose the Personal Agent Protocol, an open standard meant to govern how consumer agents interact with businesses. Bret Taylor, Sierra’s co-founder and the initiative’s leader, framed the gap bluntly: “It is kind of chaos until such a standard exists.”

The proposal would let companies distinguish a personal agent acting for a customer from an unaffiliated bot, while giving them visibility into the agent’s activity. That matters especially when agents may handle credit-card and personal information. Meta’s David Singleton likened the ambition to email: shared rails that allow different systems to communicate while preserving a degree of control.

Sierra’s formulation puts the bargain at the center: “consumers decide what access to give their personal agents, and companies set parameters for what those agents can do.” Businesses could work through existing websites and APIs or deploy their own agents, with Genesys, Instinct, Rocket and Shopify also involved.

The standard’s real test will be adoption. OpenAI and Anthropic are not yet participants, though Taylor says the point is for the protocol to be genuinely open—not another closed gate around an increasingly crowded bot economy.