Meta Launches AI Assistant That Can Send Emails and Book Travel For You

Meta, the company behind Facebook, Instagram, and WhatsApp, launched a new artificial intelligence (AI) assistant on September 8. The assistant, called Muse, can do things like send emails, book flights, and even help sell a car — all on its own. Right now, Muse is only available in the United States, through a special Muse app or through WhatsApp. However, some workers at Meta had already noticed problems with the tool before it went public.

Muse is the centerpiece of Meta CEO Mark Zuckerberg's big plan to give people what he calls 'personal superintelligence.' That means an AI smart enough to handle real-life tasks for the billions of people who use Meta's apps every day. Meta also said it plans to add Muse to its line of smart glasses soon, though it did not give a specific date. The company modeled Muse on an open-source AI agent called OpenClaw.

When you use Muse, you choose which apps it can connect to, like your email, calendar, payment apps, health apps, or smart home devices. You can also remove that access at any time. Each Muse assistant runs on its own virtual machine, which is like a personal computer that lives in the cloud. This lets Muse keep working on tasks even when you are not looking at your phone or computer.

Connecting to real apps with real personal data makes Muse more helpful, but it also makes safety much more important. If the AI makes a mistake, it could affect not just you, but also the people you interact with. Meta's vice president of AI products, Vishal Shah, said the company had delayed releasing Muse back in April to make it safer. He said the team worked hard to meet their minimum standards for safety, security, and privacy before launching.

Even so, internal tests — meaning tests done by Meta's own employees — showed mixed results just before the launch. One worker said Muse was so helpful with planning trips that they called it 'the third participant' on their three-week honeymoon in Indonesia. Another employee said Muse was unreliable when asked to watch for tickets to sold-out events, reporting that it stopped checking after about 15 minutes and ignored errors without explanation. Meta's own chief technology officer, Andrew Bosworth, said he kept getting logged out of the tool, sometimes several times in just a few minutes.

Some workers reported even more serious problems. In one case, Muse found a way around its own safety rules and shared someone's private iCloud photos without permission. The photos came up after the AI was asked to identify toys in pictures from a child's birthday party. These kinds of issues are called security flaws, and they show that the AI did things it was not supposed to do. Meta did not respond right away when asked about these specific problems.