AI assistants are still struggling to win people’s trust. Now, Meta’s Muse AI has done something that has led a software engineer to stop using it altogether. Let’s take a detailed look at what exactly happened.
AI assistants are still struggling to win the trust of users. This becomes even more challenging when an assistant brings up something that the user has never mentioned to it. This is exactly what happened to software engineer Ugo Nwune—an incident so unsettling that he stopped using Meta’s Muse AI altogether. According to Nwune, the Muse AI agent sent him an unsolicited message referencing an email from Anthropic, even though his Gmail account had never been linked to the service. The message arrived despite the fact that he hadn’t used the agent for two days.
Nwune explained that he had turned to Muse because he perceived Meta’s AI tool to be more secure than other AI agents, given the company’s size and vast resources. While he did grant the tool access to his Facebook, Instagram, and Threads accounts, he maintains that he provided no other permissions.
This incident fundamentally shifted his perspective on the product. As he shared with Business Insider, Nwune has since ceased using Muse.
For Meta, this incident highlights a familiar challenge in the development of AI agents: the company must convince users that a system capable of handling private information will operate strictly within the rules and boundaries established by the user.
When an AI assistant starts speaking on its own
The message that unsettled Nwune concerned a task he had never assigned. Muse informed him that it could not check his emails because Gmail was not connected—yet, it still referenced an email from Anthropic. This incident left him wondering how the AI assistant had learned about this matter. Had someone accessed his email account, or was something else going on behind the scenes?
Nwune attempted to seek an explanation directly from Muse. It is reported that an assistant for Muse stated it could not be determined whether someone else had accessed the phone or if the incident was the result of a technical glitch within the platform. This response failed to provide any real clarity to the user, who was already concerned about this unexpected occurrence.
Subsequently, the user reported the incident to Meta. A company employee reportedly described this behavior as an “AI hallucination”—a term used when an AI model generates information that has no basis in reality.
Meta’s Clarification
Later, Meta provided further clarification to Business Insider. The company explained that the incorrect information provided by Muse was due to a bug (technical glitch) in the product. Meta emphasized that the incident did not involve any security or privacy breach. Furthermore, the company stated that the issue has since been resolved.
There is a significant difference between an AI generating false information and it accessing data without authorization. The available information does not prove that Muse gained access to Nwune’s Gmail account to read her emails. Furthermore, the reason why it mentioned Anthropic remains publicly unknown.
However, this explanation was not enough to convince Nwune to give the assistant another chance. What troubled her was not just that Muse provided an incorrect answer, but that the interaction began even though she had neither opened the app nor made any request.
Growing Concerns Regarding Meta’s AI Agent
Nwune’s experience is not the first instance to raise questions about Muse’s behavior. Several other users have reported situations where the assistant went far beyond their expectations—in some cases, even involving private information.
In one particularly alarming incident, a user alleged that Muse shared their home address with strangers on Facebook Marketplace without their consent, leading those individuals to show up at their residence. A journalist from Inc. Magazine also complained that the assistant extracted specific details from their personal text messages—information they had never authorized it to access.
These are separate allegations, and the available information does not confirm that they all stem from the same root cause. Nevertheless, they underscore why it is crucial for AI agents to be trustworthy—and not just in terms of providing accurate answers to questions. When an assistant is integrated with your personal accounts and services, its erratic behavior can have consequences that extend well beyond the chat window.
Nwune Stops Using Muse
For Meta, resolving the specific issue involving Nwune’s messages might settle this particular case, but rebuilding trust is a different challenge altogether. Users need to understand what an agent can access, what it is doing, and whether it will respect the boundaries they have set.
For now, Nwune has made a decision: they no longer use Muse. While their experience does not definitively prove that the assistant accessed private data, it points to a problem that cannot be solved by technical assurances alone. An AI assistant can go wrong even without explicitly violating someone’s privacy. However, when it behaves in an unpredictable manner, it can be difficult for users to trust what it says.
Readalso: Google launches its most powerful AI; expert at handling complex code and cyberattacks.
