Accounts without a profile photo or real name plague social networks; perhaps even you, who read these lines, are the owner of one. We do not judge, there are many reasons not to show your face on networks and, in fact, anonymity is the pillar on which the internet has been built. However, if you thought that calling yourself ‘user84721’ and having a landscape photo protected you, researchers have just shown that accounts can be deanonymized in minutes with AI (of course).
The study. A team of researchers has published a study called “Large-scale deanonymization online with large language models” which is echoed Guardian. In it, they demonstrate how an LLM-based agent is able to compromise anonymous social media accounts with astonishing efficiency. The process consists of three steps: the LLM extracts identifying data (age, location, interests…), looks for possible matches in other users and finally reasons which are the best candidates, verifying the matches and eliminating false positives.
Minutes. This is how long it took to identify users on sites like Reddit, Hacker News, and Anthropic Interviewer Dataset participants with this method. In the image you can see how, based on a few pieces of information such as where the student studies, the approximate age, the city and the name of the dog, they achieve a match with the user’s real profile. This is a fictitious case, but in the experiment they managed to identify real users by cross-referencing information with Linkedin profiles and other platforms.
According to the researchers, LLMs allow for large-scale deanonymization of accounts, far exceeding the speed and efficiency of classical methods. They also highlight that there is not always enough information to reach a match, so everything depends on the online footprint of each user.
Consequences. Researchers warn that this use of AI could be used for problematic purposes, such as governments that want to identify activists or cybercriminals seeking to launch highly personalized attacks. In addition, it must be taken into account that the system is not infallible and there may be false positives. Speaking to The Guardian, Peter Bentley, professor of computer science at UCL, warns that “People are going to be accused of things they haven’t done.”
The end of anonymity. As we said at the beginning, the Internet has been built on the anonymity of its users, but we are experiencing a regulatory shift that pursues precisely put an end to it. We see it with the ban on social networks or the blocking of pornographic websites for minors promoted by countries such as United Kingdom, Australia, Denmark and now also Spain. These initiatives require the identification of users to be able to access certain content through video selfies, electronic ID, verification systems with AI… There are many options, What is not clear is its effectiveness.
Image | Own preparation on a background of Google DeepMind
GIPHY App Key not set. Please check settings