The new Muse AI agent from Meta has one main purpose: to do something that chatbots can’t do – act on your behalf. It can surf the web, send e-mail messages, fill in forms, do some online shopping, book travel, and more – and do so without you having to do everything.
That sounds convenient. As millions of individuals begin to grant an AI agent access to websites, however, there is a larger question that is arising: What is going to happen when the internet becomes a space for millions of AI users, in addition to humans?
The Internet is designed for People, Not AI Assistants
Most sites are built based on human behavior. Someone could be looking for a product, emailing, making one booking, or filling a form. An AI agent could easily do those actions many times more quickly and across many services.
This leaves us with another issue. When many users submit requests simultaneously – say, when thousands of customers phone agents to request the best concert tickets or restaurant reservations, or the limited products on sale – websites can suddenly see a huge automated demand.
It is already apparent in the overall trend. Bad bot traffic rose 124 percent, and AI traffic rose 82.3 percent from July 2025 to June 2026, according to DataDome’s analysis of over one trillion web requests over the past year. The firm also discovered that automated traffic accounted for around 26.5% of the traffic that it analysed.
A Personal Agent Can Also Become a Powerful Bot
What if you asked Muse to search for a popular product for the best price? The agent can access multiple websites and interact with sellers in place of making five trips to the stores. Through its virtual computer, Muse can do tasks like making appointments, filling out forms, and interacting with customers for customer service.
If that’s the case, multiply that behavior by millions of users and businesses, and they may find themselves having to differentiate between actual customers and millions of automated interactions.
However, there are already some indications of the complexity that this will entail. The volume of AI-agent targeting of login endpoints grew by 735.8% in the first half of 2026, which points to a shift in automated systems beyond mere webpage reading to engage with critical online services, says DataDome.
It may lead to an increase in CAPTCHAs, more stringent bot detection, and websites becoming more difficult to use for regular users.
Muse’s Biggest Challenge May Be Trust
There’s another problem: via an AI agent, you’re opening a far bigger security risk than with a regular chatbot.
The controls are supposed to ensure that users have control over what the agent can access, and Meta says Muse is based on a dedicated Muse Secure VM.
However, it’s researchers who have already discovered problems. Security researcher Patrick Wardle found that the Mac version also had a security flaw that could enable a hacker who was already running code on the Mac to intercept the use of Muse. Meta then issued a patch for the bug.
As another vulnerability was found that could have revealed sensitive data from its virtual machines, the sources said that Meta beefed up Muse’s safety messages. While these incidents are not a reflection of an inherently unsafe nature of Muse, they do underscore the need for a new approach to security when it comes to autonomous AI. The more that an AI performs, the more it’s worth to an attacker to gain access to it.
The Web May Have to Change With AI Agents
The problem with Muse is not just that one. OpenAI has also launched its own agent, and other businesses are working on developing their own.
That implies the transition to agent-driven compute isn’t about just one Meta product.
Websites may eventually require alternative methods to let AI agents know who they are, how to use them, and how to interact with the sites without overwhelming them. Also, businesses can have to redesign systems which are currently built with the premise that all visitors are human.
For users, it seems quite clear: less clicking, less paperwork and more things done for them. However, the changeover might be messy. Muse is perhaps a glimpse into a world of the internet where the biggest change is not one in which AI can answer a question, but one in which it can begin to act upon a question.
