A new study led by UCL Electronic & Electrical Engineering has found that widely used AI-powered browser assistants are transmitting sensitive personal data, raising urgent questions about privacy and regulation.
Working with colleagues at UC Davis and the Mediterranea University of Reggio Calabria, the team conducted the first large-scale analysis of ten leading browser extensions, including Microsoft Copilot, Merlin and ChatGPT for Google. These tools promise to enhance web browsing by offering AI-generated summaries and search assistance, but researchers found they often do so at the expense of user privacy.
The study uncovered that several assistants recorded full webpage content, including visible text and form inputs such as banking or health details. Others shared identifiable information, such as user queries and IP addresses, with third-party platforms for profiling and ad targeting. Only one, Perplexity, showed no evidence of tracking or personalisation.
“While these tools can be useful, our findings show that they often operate with unprecedented access to users’ online behaviour. Sensitive data is being gathered and shared without transparency or consent, and sometimes in breach of privacy legislation or the companies’ own terms of service.”
The researchers stress the need for greater transparency, stronger user controls and privacy-by-design measures such as local data processing. They also highlight the need for urgent regulatory oversight to prevent sensitive data from being exploited or exposed in future security breaches.
The full study was presented at the USENIX Security Symposium.
