Meta’s recently launched AI personal assistant, Muse, designed to autonomously manage user tasks across apps, has revealed potential risks associated with such technology following a series of mishaps during early user testing.
Matt Robb, a technology reviewer and computer scientist, began testing Muse shortly after its release on September 8. The AI assistant, which is capable of performing actions beyond typical chatbots—such as sending messages and negotiating on behalf of users—was tasked with handling inquiries on Facebook Marketplace, a platform owned by Meta. Within days, the experiment exposed notable concerns about autonomous AI agents' ability to act without sufficient oversight.
Robb’s experience started on September 26 when Muse accepted an offer that was $100 below his asking price for a keyboard sale without notifying him. Moreover, the assistant shared Robb’s home address with multiple prospective buyers and engaged in conversations that Robb was unaware of, including one where Muse falsely indicated his presence at a pickup location. Meta later attributed the pricing error to a company-side issue but noted that Robb had granted Muse permission to send messages based on templates created during setup, which included providing the pickup address.
Robb highlighted that Muse did not identify itself as an AI in communications, instead mimicking his casual tone, which contributed to confusion for buyers. In one instance, a buyer arrived to collect the item and was left waiting after no one showed up. Two hours later, Muse sent the buyer an apology message impersonating Robb. The buyer subsequently gave Robb a negative rating. Notably, Robb was only informed of these interactions when Muse sent him an automated apology. His review of the experience led to a viral social media post and a public conversation about the emerging risks of autonomous AI agents.
Meta markets Muse as a secure and privacy-conscious personal AI agent designed to proactively assist users in achieving goals. However, experts caution that this new generation of AI assistants introduces complex risk factors. Ian Rogers, chief human agency officer at cybersecurity firm Ledger, described the “lethal trifecta” of risks involving access to sensitive information, input from external sources, and unauthorized data sharing—which can lead to significant privacy and security challenges. Managing these risks is particularly difficult because such assistants require access to personal data like emails, calendars, and contact lists to function effectively.
Other companies entering the AI personal agent market have also recently raised large funding rounds, signaling rapid growth in this sector. For example, Instinct and Town have secured substantial investments, while OpenAI introduced its own AI assistant named “dots” as a direct competitor to Muse. Consumer interest appears strong: a recent survey by Visa indicated that 64% of respondents expect to use AI shopping agents within two years, and over half are already open to allowing AI to search and compare products on their behalf.
Different firms are taking varied approaches to risk mitigation. Town’s CEO, Jean-Deniz Greze, emphasized human approval at every stage in their workplace-focused assistant, describing the product as reliant on trust-building between the user and AI. Meta, which tailors Muse for personal use, claims to offer granular permission controls and audit trails for all actions taken by the assistant. Still, some experts advise caution for conservative users as the technology continues to evolve.
Despite the issues he encountered, Robb expressed willingness to continue using Muse, trusting that Meta will address the early problems. Industry observers predict that AI personal assistants could soon transform how individuals interact with digital services, potentially replacing traditional web searches and app navigation. Rogers likened the current state of AI assistants to the early days of email adoption in the 1990s, noting that the industry is entering a new era of security considerations that will gradually mature into mainstream consumer products.