For most of the internet era, granting a website access to your digital life was a relatively benign thing to do. You might allow a budgeting app to track your spending, give a travel app access to your calendar, or let an email service scan your inbox to filter spam. The software could organize important information and make recommendations, but for the most part, a human was still required to get a result.
With the recent release of Muse from Meta Platforms and its astounding acceptance among consumers, artificial intelligence (AI) is beginning to change this relationship entirely. Until now, most consumer AI apps were used to answer questions and perform deep research. But Meta’s new personal agent is designed to do a lot more with minimal human intervention and oversight, begging the question of who is going to keep the agents in check.
For decades, the software that cybersecurity companies have built has primarily been designed to prevent other humans from gaining unauthorized access to our digital lives. But agentic AI introduces an entirely different challenge, as we are the ones voluntarily handing the software the keys.
Identity has long been one of the foundational concepts in online security. Before someone can access their bank account, personal health data, or make a purchase, the website generally wants to establish that they are human and that they are authorized. This works seamlessly when the person logging in and the person requesting the action are one in the same.
Agentic AI breaks that logic, and the consequence for making a mistake could be considerable. For example, an agent that has been asked to research a stock may later decide to go a step further and place a trade on the owner’s behalf. Whether that trade makes money or not, it could have tremendous—and typically irreversible—consequences.
To its credit, Meta says that it has built a dedicated security architecture around its Muse agents with isolated storage for data and login credentials. But therein lies the dilemma. Can the company charged with building and monetizing the AI agent also be trusted to fully secure it, or will consumers need their own independent security layer to limit their agent’s access and provide oversight.
Historically, selling cybersecurity tools to consumers has proven difficult. But with the rise of antivirus programs, identity-theft monitoring, and password managers, perhaps the rise of AI agents targeting consumers creates an entirely new opportunity for software providers and investors.
As AI agents become more widely adopted and more capable, consumers may eventually decide to shift their focus from how novel they are to finding a simple way to set limits and maintain control. And if the current excitement and staggering adoption rates hold, Muse and similar agents could create an entirely new market for the companies that figure out how to solve it.
Related: AI Is Creating a New Cybersecurity Problem—and Potential Acquisition Targets
