Meta’s new AI agent Muse could make everyday online tasks easier, but experts warn that widespread use of such tools could create a new wave of problems across the internet.

Muse is designed to perform tasks on behalf of users, including cancelling subscriptions, contacting customer service, buying groceries and making bookings. It can access a user’s Gmail, calendar and computer, and Meta plans to add phone-call capabilities.

The launch comes as major technology companies increasingly develop AI agents capable of acting independently online. OpenAI has also announced its own agent.

Experts say the technology could overwhelm websites and services as millions of AI agents begin carrying out tasks simultaneously.

One major concern is online bookings. AI agents could compete with humans for concert tickets, restaurant reservations and government appointments, potentially allowing users to reserve multiple slots with little effort.

Calli Schroeder, director of the AI and Human Rights Program at the Electronic Privacy Information Center, said the problems caused by bots could escalate significantly as AI agents become widespread.

The impact could also extend to businesses. While a human customer may send only a handful of enquiries, an AI agent could send hundreds in a day. Websites that rely on advertising and human visitors are already struggling as search engines and chatbots increasingly provide information without sending users to the original sites.

A separate analysis found bot traffic increased 124% in the year to June 2026, adding to concerns about the strain AI agents could place on online infrastructure.

Meta says Muse has safeguards to prevent such behaviour. The company says the agent asks for permission before making purchases and is designed to act like a “reasonable, honest person” rather than repeatedly refreshing pages or buying up resources.

There are also concerns about whether AI agents will always act in their users’ interests.

An agent planning a holiday, for example, could recommend flights or hotels based on factors other than price or quality. Similarly, music recommendations could potentially be influenced by commercial relationships.

Meta says users remain in control of Muse and that the system has protections designed to safeguard their interests.

Privacy and security remain another major concern because agents need access to highly sensitive information to function effectively.

Muse can be connected to Gmail and other services, giving it access to emails, documents, financial information and personal conversations. Meta says the data collected by Muse is not connected to its advertising systems and that safeguards are in place to prevent access to passwords and payment information.

However, Muse data is used to train Meta’s AI models by default, although users can opt out. Security researcher Patrick Wardle also identified a critical vulnerability in the app shortly after its launch, which Meta subsequently fixed.

Researchers say the solution will require new rules and infrastructure designed specifically for AI agents.

Ramesh Raskar, an associate professor at the Massachusetts Institute of Technology, compared AI models to engines and AI agents to cars, arguing that regulators need to establish “rules of the road” governing how agents interact with websites and services.

The technology industry is already developing standards that could allow AI agents to interact more directly with online services, while some companies are exploring ways to charge AI systems for accessing websites.

Some platforms have also begun restricting AI-driven bookings, while the UK has changed its driving-test booking system partly to combat automated bots.

AI agents could eventually make the internet more efficient, researchers say. But until the web is redesigned to accommodate them, their rapid expansion could bring a period of greater competition, disruption and privacy risks for ordinary users.