Artificial intelligence can now create detailed personal profiles by analyzing the ads users see, without direct access to personal data. Researchers have demonstrated that the patterns of advertisements displayed to individuals can reveal significant personal traits and information, according to Tech Radar.
A study by UNSW Sydney and QUT involving over 435,000 Facebook ads from 891 Australians revealed that large language models can infer a user's gender, age, education, employment, political leanings, and economic status solely from the ads they are exposed to. This AI-driven profiling is significantly faster and cheaper than human analysis, with short browsing sessions providing sufficient data for accurate inference. The most probable attack vector involves malicious browser extensions, which already possess the necessary permissions to read web page content and can discreetly collect ad data. This method is particularly concerning due to its stealth, as the extensions continue to function normally, making detection difficult.
The advertising platforms themselves are unaware of this repurposing of their delivery systems for surveillance. While VPNs offer no protection against this form of profiling, individuals are urged to be cautious with browser extensions and adjust privacy settings. The researchers advocate for updated privacy laws that address inferred data, not just collected data, to protect users' online privacy.
Source: Tech Radar
