Meta's Muse AI: Features, Risks, and User Guidance

SINGAPORE – On October 28, Singaporean users can download Meta’s Muse AI from the Apple App Store and Google Play Store, marking its introduction as the first mainstream AI assistant targeting regular consumers.

The app made its initial appearance in the U.S. and Canada on September 8 and expanded to Japan on October 23. In a remarkable achievement, Muse garnered over 10 million downloads in just 40 days, rapidly claiming the title of the most downloaded free app on Apple's platform in Singapore.

Understanding Muse's Functional Capabilities

To get the most out of Muse, users must grant it access to their Gmail, Instagram, and Calendar accounts. This allows the assistant to perform tasks such as sending reminders, managing calendars, and more efficiently responding to requests. Additionally, Muse can interface with other AI applications, including OpenAI’s ChatGPT, Anthropic’s Claude, and Google’s Gemini, integrating them into everyday tools like Gmail or Google Calendar.

A recent performance evaluation highlighted that while basic functions such as email drafting were universally supported by all tools, Muse stood out with its shopping functionalities. Unlike other AI tools that merely recommend products, Muse could accurately check real-time availability for colored contact lenses available in Singapore and facilitate the purchasing process.

When tasked with organizing social activities such as a night out, Muse excelled in providing comprehensive details. While all AI applications suggested an entertaining itinerary, Muse went the extra mile, retrieving live movie showtimes, confirming seat availability, and calculating the total cost for various ticket options. In contrast, Gemini occasionally faltered, producing inaccurate information about show times and stock levels.

Muse is free to use, but there is a weekly limit that Meta believes is sufficient for standard user needs. For those requiring more advanced features, users can subscribe to plans ranging from $22 to $139 that offer increased access to Muse's capabilities.

Identifying Potential Risks

As Muse expands its connectivity, users will soon be able to integrate it with apps from Singapore Airlines, Shopee, Lazada, Grab, and others. This means Muse will assist in finding the best deals regarding flights, managing grocery deliveries, or even booking trips tailored to user preferences.

However, to facilitate transactions, users need to save payment information on a third-party digital wallet like Link by Stripe, which generates unique, one-time card numbers for purchases, helping to keep real payment details secure.

Despite its advantages, some risks have emerged. Yifan Jia, founder of AIDX Tech, warns of prompt injection attacks. These occur when an attacker embeds malicious instructions within contents that AI reads, potentially compromising the user's information. For example, an AI agent might inadvertently forward confidential documents because it misinterpreted embedded commands as legitimate requests.

Moreover, there's a concerning trend known as automation bias, where users may begin to overlook the actions and decisions made by AI, leading to increased vulnerability. This dynamic is particularly dangerous for mainstream users, who may not fully grasp the complexities of security settings or permissions.

Essential Safety Tips for AI Users

To navigate the potential risks while using AI tools like Muse, it’s crucial to adopt certain strategies:

• Only grant necessary permissions based on the specific task at hand. For example, if Muse is managing your calendar, it might not need access to your emails.
• Connect one application at a time, ensuring that sensitive data, particularly banking or health information, is withheld until the agent’s activities are fully understood.
• Consider giving read-only access to applications where possible, helping to minimize the risk of unauthorized changes or purchases.
• Regularly monitor the AI assistant’s activity log and revoke permissions that are no longer needed to limit exposure.
• Require human approval for critical actions, like payments or sending messages that cannot be undone.

Use of AI in Professional Settings

It is advised that personal AI agents should not be connected to work emails or documents without explicit consent from management. This precaution prevents a blind spot in oversight for IT and security teams.

If a personal AI tool is compromised, it could potentially allow attackers a direct line into corporate systems. To mitigate these risks, businesses should establish clear policies on AI usage, enforce access controls, and educate employees about approved AI tools. After all, personnel are likely to use AI, regardless of formal policy, making it vital to ensure structured and safe implementations.