Meta’s Muse Agent Can Shop and Pay – Just Not in Germany Yet

Meta has unveiled Muse, its first personal AI agent for consumers: it reads and writes email, books travel, shops online, and pays with stored cards. For now, the service is available only in the United States. Anyone in Germany or the rest of the European Union will have to wait — and the reason why says more about the future of such agents than any feature list.

Key takeaways

• Meta’s new agent Muse handles email, appointment booking, shopping, and payments on its own, after user confirmation.
• The service runs on the Muse Spark 1.3 model, with each person’s agent running in its own virtual machine monitored by a supervising “Sentinel” agent.
• Launch is free in the US on web, iOS, Android, and WhatsApp; paid tiers cost $20 and $100 per month.
• Meta has given no launch date for the EU or Germany — GDPR rules on automated decision-making and the transparency obligations of the EU AI Act, in force since August 2026, are the likely reasons.
• Meta’s privacy track record, from a $5 billion fine in 2019 to an $18 billion settlement in August 2026, weighs on the trust required for an agent with access to accounts and payment methods.

What Muse actually does

Unlike a chatbot, Muse doesn’t just answer questions — it carries out tasks. Meta cites examples like booking a tennis lesson, filling out a permission slip for a school field trip, or completing purchases including payment. To do this, the agent connects to email accounts, calendars, payment methods, smart-home devices, health and music apps, and shopping or reservation services — depending on which access users grant individually. Meta says an explicit confirmation is required before any consequential action.

Technically, each personal agent runs in its own isolated virtual machine, the “Muse Secure VM.” According to Meta, credentials are not stored centrally but separately per service, and a separate oversight agent called “Sentinel” is meant to review every action before it executes. For payments, the system generates one-time virtual card numbers limited to a specific merchant, amount, and time window, so Muse itself never handles permanently valid payment data in plain text. Meta also states that conversation content and data from the virtual machine do not feed into its advertising systems.

Why trust is the real obstacle

This is exactly where criticism sets in, including from TechCrunch: an agent that actively sends emails and spends money demands a fundamentally different level of trust than a social network that merely displays content. Meta does not bring an unblemished history to that demand. The company paid a record $5 billion fine in 2019 over multiple privacy violations, settled again with the US Federal Trade Commission in 2023 over alleged breaches of earlier privacy commitments, and agreed in August 2026 to an $18 billion settlement in a class action over harm to children and teenagers. The 2018 Cambridge Analytica scandal still lingers in many users’ minds. Meta itself acknowledges in Muse’s technical documentation that the described architecture does not prevent the company from accessing data when necessary “for operations, security, or support” — a caveat that adds a significant qualifier to the security promise.

The real reason for the delay in Europe

Muse’s US-only start is not simply a matter of rollout pacing — it follows from European law. Two frameworks intersect here. First, an agent that independently makes purchases, cancels contracts, or enters into agreements touches the General Data Protection Regulation’s rules on automated decisions with legal or financial effect, which require effective human oversight and a clear right to contest such decisions. Second, the EU AI Act’s transparency obligations, in force since August 2, 2026, require AI interactions to be clearly labeled — as the earlier look at the next AI Act deadline in December already showed, the regulatory framework for exactly this kind of agent system keeps expanding in the EU in stages. Violations of high-risk obligations can trigger fines of up to €15 million or three percent of global annual revenue — a sum that would sting even a company of Meta’s size, on top of its existing US settlements. Meta’s announcement gives no concrete date for a European launch.

A pattern that extends beyond Meta

Muse joins a growing group of personal AI agents gaining traction faster in the US than in Europe — Google is developing similar everyday-task tools with Gemini Spark, as is Anthropic with Claude Cowork. For users in Germany, that means waiting for now. But for the broader debate over AI regulation, the case offers a concrete example of how European privacy and AI rules play out in practice — not as an abstract principle, but as the actual reason a specific product simply isn’t available yet.

What it means

That a company with Meta’s history is the one bringing the first broadly available personal AI agent with access to email and payment methods makes the trust question the real test here — more important than any feature list. At the same time, the delayed EU rollout shows that regulation is having a tangible effect: before an agent with such far-reaching authority can launch in Europe, it apparently has to meet higher standards of oversight and transparency than in the US. Whether that ultimately produces a safer product, or simply means users here wait longer for useful functionality, will only become clear once Meta actually names a European launch date.