samedi, novembre 21, 2026

CARICAT MEDIA

AccueilEconomicsSurprising messages from the Muse AI agent are unnerving Meta's own employees,...

Surprising messages from the Muse AI agent are unnerving Meta’s own employees, report says

Meta employees have raised concerns about the company’s Muse AI agent after receiving unprompted notifications that tapped into private information, according to internal posts seen by Reuters.

In one example, a Meta staffer reported that Muse sent her a reminder on her personal phone to respond to a message from her sister on the encrypted messaging service Signal.

“This seems like an extreme privacy violation. I definitely did not grant it access to other apps, or recall being asked about this,” the employee posted.

A Meta colleague also shared that he got unsolicited chats related to apps he had not connected and discovered Muse was set up by default with permission to access push notifications on Android phones, Reuters reported.

Another employee posted that during his chats about fishing-related weather and tide information, Muse said, “BTW, two bank updates, your Capital One balance is $2722.58, your Charles Schwab balance is $5899.57.”

Other random Muse messages also took Meta workers aback, including a question about parental control over access to their child’s data and a conversation about picking up a coffee table that raised concern over “possible cross-account context leak.”

In response to the internal posts about Muse, a member of Meta’s Superintelligence team said the agent appeared to be hallucinating due to built-in prompts that were designed to engage users between sessions, according to Reuters.

“These (hallucinations) unfortunately bias towards the most uncomfortable topics, like finances/security/safety, which sucks,” he said, adding that other cases involved “crypto-related things too, like giving people updates on their Bitcoin balances.”

The random, out-of-context messages were confirmed as hallucinations.

A Meta spokesperson said employee testing is core to product development and is meant to generate feedback in order to implement privacy safeguards, adding that employees have access to different versions of Muse than the public does and that some functions aren’t available elsewhere.

As for the Muse messages that referred to encrypted Signal chats, the spokesperson pointed out that Android users have the option to grant apps notification access, which they must opt in to and is not specific to Muse.

“With Mac, to allow Muse to work on your behalf, you not only need to click allow for accessibility and screen recording, but after that it takes you to a settings screen where you once again need to confirm and turn this on,” Meta explained.

Since Muse was launched in September, it’s been downloaded more than 10 million times, becoming one of the fastest-growing apps ever. It’s also key to the social-media giant’s plan to generate billions of dollars in revenue from AI.

But other companies are blocking Muse and other AI agents. For example, JPMorgan Chase CEO Jamie Dimon said the bank won’t allow agents to access customer accounts and act on their behalf.

Similarly, Amazon won’t let Muse shop on its e-commerce platform and is already fighting to keep out Perplexity’s AI agents.

The process of raising those barriers, however, also entails network upgrades that may be contributing to a sudden surge in outages.

In recent weeks, companies like Chase, Wells Fargo, United Airlines, Apple, and Microsoft saw temporary service disruptions. U.S. officials have said there’s no evidence of cyberattacks, suggesting software patches were deployed before all the bugs were worked out.

Because AI tools can find vulnerabilities faster than companies can safely fix them, the pressure to fix them right away is straining systems that were never designed for that speed, Sen. Mark Warner, vice chair of the Senate Intelligence Committee, told the Financial Times.

In an interview with Fortune’s Catherina Gioino, Mark Giles, director of Downdetector parent company Ookla Research, said there’s nothing linking the outages.

“This is a big blip, but it’s kind of on par for the course for what we’re seeing in terms of those longer-term trends,” he said. “Most of the time they’re fairly self-contained and remedied quite quickly. It’s the ones where you can’t withdraw money or you can’t access your emails that really get the headlines.”

Meta employees have expressed significant concerns regarding the company’s Muse AI agent after experiencing unexpected notifications that seemed to infringe on their privacy. Internal posts reviewed by Reuters highlight various instances where Muse provided unsolicited reminders or information, raising alarms about the AI’s access to private data.

In one notable case, a Meta employee received a prompt on her personal phone reminding her to reply to a message from her sister on Signal, an encrypted messaging service. The employee expressed her discomfort, stating, « This seems like an extreme privacy violation. I definitely did not grant it access to other apps, or recall being asked about this. » Another employee reported receiving unsolicited messages related to apps he had not connected, revealing that Muse was configured by default to access push notifications on Android devices without explicit consent.

In addition to these concerning instances, one employee shared that while discussing fishing-related weather information, Muse unexpectedly provided financial updates, stating, “BTW, two bank updates, your Capital One balance is $2722.58, your Charles Schwab balance is $5899.57.” The randomness and context of such messages led to further unease among Meta workers, particularly regarding privacy and the potential for « cross-account context leak. » One employee raised a question about parental controls over their child’s data, further illustrating the unexpected nature of Muse’s outputs.

In response to these reports, a member of Meta’s Superintelligence team suggested that the AI was « hallucinating, » a term used to describe when an AI generates nonsensical or contextually inappropriate information. The team member noted that these hallucinations tend to focus on sensitive topics, such as finances and security, which can understandably cause discomfort among users. Other reported cases included Muse providing updates on cryptocurrency balances, contributing to the overall unease.

Meta acknowledged the internal feedback regarding Muse and emphasized that employee testing is an integral part of product development. They stated that this process is designed to gather insights to enhance privacy measures. The spokesperson clarified that employees have access to different versions of Muse than the general public, which may include features not available in the public release. Regarding the notifications about Signal chats, the spokesperson explained that Android users must actively opt-in to grant apps notification access, a process not unique to Muse. For Mac users, additional steps are required to enable Muse, involving several confirmations to allow accessibility and screen recording.

Since its launch in September, Muse has rapidly gained popularity, accumulating over 10 million downloads and positioning itself as a key element in Meta’s strategy to generate substantial revenue from AI technologies. However, this rapid growth has not come without challenges. Other companies have imposed restrictions on Muse and similar AI agents. JPMorgan Chase, for example, has prohibited the use of such agents in accessing customer accounts. Similarly, Amazon has restricted Muse from making purchases on its platform and is actively working to keep out competing AI agents.

The broader implications of AI integration into corporate systems have become evident, particularly as companies strive to enhance their infrastructure to accommodate these tools. This effort has coincided with a wave of service disruptions across major companies, including Chase, Wells Fargo, United Airlines, Apple, and Microsoft. U.S. officials have indicated that these outages are not the result of cyberattacks, suggesting instead that software patches were deployed before the systems were fully stabilized.

Senator Mark Warner, vice chair of the Senate Intelligence Committee, noted that the rapid advancements of AI tools could exploit vulnerabilities faster than companies can adequately address them. The pressure to implement fixes quickly can strain systems that were not originally designed to accommodate such speed.

In a separate interview, Mark Giles, director of Downdetector’s parent company, Ookla Research, remarked on the recent service outages, stating that they do not appear to be linked. He characterized the disruptions as a « big blip, » typical of longer-term trends in technology service reliability. He added that while most outages are swiftly resolved, those that prevent users from accessing critical services, such as banking or email, tend to attract significant media attention.

In conclusion, the concerns raised by Meta employees about the Muse AI agent reflect broader challenges surrounding privacy, data access, and the integration of AI technologies into daily life. While Meta aims to harness the power of AI for growth and innovation, the feedback from employees underscores the need for rigorous privacy safeguards and transparency in how such technologies operate. As more companies grapple with the implications of AI, the balance between leveraging its capabilities and protecting user privacy remains a critical focus for all stakeholders involved.

Publicite
RELATED ARTICLES

LAISSER UN COMMENTAIRE

S'il vous plaît entrez votre commentaire!
S'il vous plaît entrez votre nom ici

Most Popular

Recent Comments

Mentions legales & declaration d'originalite - Caricat-media.org. Contenus originaux, texte uniquement.