Can AI agents protect our privacy? 

AI is rapidly transforming the Internet: In 2025, retailers, news publications and marketing agencies saw a drop in web traffic of 20 to 40 percent. ChatGPT and other AI agents are browsing the web for us. AI agents can book our flights. They can browse news sites, dashboards and public records to compile research memos. They may even order our groceries. While our AI agents engage in these activities, they can also expose our personal data.  

To be sure, that trend is not new. Since the advent of online advertising, websites routinely hand peoples’ data to ad networks and data brokers, and from there it can get into the hands of the government. However, in the age of the AI-driven web, the risks to our privacy are even greater. We may ask ChatGPT for financial advice. It may also know about our ailments and mental health. Our AI agents know a great deal about us. 

Recently, OpenAI and Microsoft announced the rollout of ads within ChatGPT and Copilot. Serving ads is the most common business model on the web today. OpenAI and Microsoft are almost certainly not the last companies that will introduce ads in their AI agents.

Commonly, ads are targeted to users based on the websites they previously visited, where they live, their inferred interests, and many other types of data that ad networks collect and aggregate into user profiles. Against this background, trying to transfer the existing ecosystem of surveillance-capitalism to the AI-driven web is not a good idea. Rather, at this point, we should use the opportunity to meaningfully improve privacy. As we are building the foundations of a new web........

© The Hill