Can Meta’s $145bn AI bet succeed if Muse fails at basic travel bookings?

Meta’s newly launched AI assistant, Muse, demonstrated significant limitations when tasked with booking holiday travel, according to a report by Gizmodo. The agent, which Meta positions as a round-the-clock personal assistant, struggled with what should be routine interactions. This performance gap highlights the operational risks businesses face when deploying autonomous systems that require access to sensitive personal data and financial information.

How capable is Meta’s Muse agent in practice?

Testing revealed that Muse could not reliably handle complex, long-running tasks without human intervention, contrary to the promise of autonomous assistance. The agent got hung up on basic steps, suggesting that the technology is not yet ready for unsupervised use in critical workflows. Gizmodo’s evaluation found the experience discouraging, indicating that current AI agents still require significant monitoring and sign-off from users to prevent errors or unexpected actions during multi-step processes.

What is the business context for this product?

Meta is investing heavily in AI agents to justify its massive technology spend, which is projected to reach $145 billion this year. CEO Mark Zuckerberg described Muse as the centerpiece of the company’s current building efforts during the Meta Connect conference. Despite this strategic importance, the agent’s reputation remains mixed, especially in the wake of security incidents involving other AI platforms. The company relies on user adoption to validate this expenditure, even as competitors like OpenAI launch visually similar agents called dots.

Why does this matter for enterprise risk?

Autonomous agents require broad access to personal data to function effectively, creating significant privacy and liability trade-offs for organizations. Meta’s own signup page warns that the agent can occasionally make mistakes, a disclaimer that underscores the inherent risk of delegating tasks to imperfect systems. For businesses, this means that integrating such tools into customer service or internal operations could lead to errors that damage brand trust or expose sensitive information, requiring robust oversight and clear liability frameworks.

What should leaders do next?

Business leaders should approach AI agent adoption with caution, recognizing that current tools are not fully autonomous. Organizations should maintain human oversight for critical tasks and carefully evaluate the privacy implications of granting agents access to corporate or customer data. It is essential to test these systems thoroughly in controlled environments before scaling them to production, ensuring that any errors or unexpected behaviors are caught early. Monitoring vendor updates and security advisories will also help manage evolving risks.

Background: Sensitive personal data

Classifying sensitive personal data allows teams to apply specific controls to high-risk assets. Without this distinction, resources are wasted protecting low-value logs while critical identity information remains exposed to unauthorized access and misuse.

Read the full guide: Why Classifying Sensitive Personal Data Prevents Security Failures

What to do and how to stay safe: Meta

• Review data access permissions for any AI agents integrated into your workflow to limit exposure.
• Implement human-in-the-loop protocols for critical tasks to catch errors before they impact operations.
• Monitor vendor communications for updates on agent capabilities and known limitations.
• Conduct regular audits of AI interactions to ensure compliance with privacy and security standards.

Step-by-step guide: How AI Risk Management Works: The Hidden Failure Points

General security guidance from the Sandbox Dispatch newsroom. It is not confirmed advice from the organisations named in this story.

Frequently asked questions

Can Meta’s Muse agent book travel independently?

No, testing showed that Muse struggles with basic travel booking tasks and requires human oversight.

How much is Meta spending on AI this year?

Meta is on track to spend as much as $145 billion on its AI buildout this year.

Are there other companies launching similar AI agents?

Yes, OpenAI recently unveiled its own AI agents, called dots, which resemble Muse aesthetically and functionally.