Meta’s new Muse AI agent is probably one of the fastest-moving stories in artificial intelligence at the moment, with millions of downloads, surging investor interest and growing adoption in online commerce. But less than three weeks after launch, the technology is also facing its first major tests around privacy, security and how much control AI agents should have over the internet.
The latest controversy surrounds Muse’s ability to make phone calls on behalf of users. Reuters reported that Meta tested a “human concierge” system where contractors quietly handled some calls initiated through Muse. The experiment was reportedly designed partly to improve call success rates after businesses sometimes hung up when they realized they were speaking to an AI.
Internal posts reviewed by Reuters showed that some Meta employees raised concerns that users could unknowingly share sensitive information with contractors. Meta subsequently rolled back the human-assisted feature and said any future public rollout would include appropriate disclosures and more privacy protections.
Muse is quickly becoming much bigger than a chatbot
Muse is Meta’s attempt to turn AI from something people talk to into something that actively works for them. The agent can browse websites, fill out forms, send emails, book travel, negotiate bills and make purchases. Meta says Muse operates inside its own dedicated Secure VM and requires approval before sensitive actions like purchases or sending emails.
(Source: Meta)
Early adoption has been enormous. Muse recorded about 2.8 million downloads in its first 12 days, while Meta shares have climbed by more than 20% since its release, adding roughly $200 billion in market value.
Meta’s stock price over the past month (Source: CoinCodex)
Muse is also becoming the center of a battle over agentic commerce. Amazon blocked the assistant from shopping on its website, and argued that unauthorized third-party agents should not make purchases without the platform’s approval. Shopify moved in the opposite direction by supporting Muse through Shop Pay, while PayPal is also integrating with the agent.
Security is another concern. Meta already patched a vulnerability in the Mac version of Muse after researcher Patrick Wardle proved that malicious software already running on a machine could potentially hijack the agent’s access. Meta made it clear that the flaw required an existing local compromise rather than providing attackers with a remote entry point.
Muse’s quick rise therefore is more than another AI product launch. It is becoming an early real-world test of what happens when AI agents move beyond generating answers and begin making calls, spending money and interacting with businesses on behalf of millions of people.