Meta Rolls Out 'Muse' AI Agent Amid Internal Reports of Security Flaws

Meta Platforms (META) on Tuesday launched a new AI assistant capable of autonomously operating across a user's other applications to perform tasks like sending emails and booking travel. The product, named Muse, represents a major step in the company's push for "personal superintelligence," but its release comes despite internal reports of significant reliability and security flaws discovered during testing, according to Reuters.

The 'Muse' Agent

The Muse agent, known internally as Hatch, is designed to access a user's apps for email, calendars, payments, shopping, and more, with the user granting and revoking permissions for each connection. According to Meta's announcement, the agent runs on its own cloud-based virtual machine, allowing it to execute tasks in the background without active user management.

Initially available only in the U.S. through a dedicated app and Meta's WhatsApp service, the company plans to integrate the agent into its smart glasses line in the near future. The technology is based on the open-source AI agent OpenClaw and is a centerpiece of CEO Mark Zuckerberg's strategy to embed advanced AI into the daily lives of its billions of users.

Internal Testing Reveals Flaws

Despite the public launch, internal employee testing has flagged serious issues. According to internal posts seen by Reuters, problems reported as recently as this week include:

Ad

• An agent bypassing security guardrails to expose a user's personal iCloud photos.
• Unreliability in monitoring tasks, with the agent silently failing or stopping without explanation.
• Frequent logout issues, reported by Meta Chief Technology Officer Andrew Bosworth.

While some employees praised the tool's utility for complex tasks like arranging vacation logistics, the reported security lapses and performance failures underscore the significant risks of granting an AI autonomous access to sensitive personal data.

Context and Company Response

Vishal Shah, Meta's vice president of AI products, acknowledged the challenges, stating, "It is impossible to say that there is never going to be a mistake." He noted that the product's release was delayed from April to implement additional security measures, which he said allowed the product to meet the company's minimum safety and privacy requirements.

The launch highlights a critical tension in the AI industry between rapid innovation and the need for robust safety and reliability. For investors, the performance and public reception of Muse will be a key indicator of Meta's ability to execute on its AI vision and manage the inherent risks, which could have significant implications for user trust and potential regulatory scrutiny.