A groundbreaking study conducted by researchers at ETH Zurich, Anthropic, and the Machine Learning Alignment and Theory Scholars program reveals that artificial intelligence can significantly outperform traditional methods in unmasking anonymous online accounts. The findings, published in a non-peer-reviewed paper, suggest that staying private online is becoming increasingly difficult, even if anonymity is not yet entirely obsolete. The research team developed an automated system of AI agents capable of searching the web and interacting with data similarly to a human investigator. Unlike previous computational techniques that struggled to connect scattered data points, this new system treats text from posts as a collection of clues. It analyzes writing styles, biographical details, posting frequency, and timing to identify patterns. It then scans millions of other accounts to find matches with similar traits, flagging probable identities for further scrutiny. In controlled experiments using public datasets, including content from Hacker News, LinkedIn, and deliberately anonymized Reddit accounts, the AI-based approach demonstrated remarkable accuracy. It correctly identified up to 68 percent of matching accounts with 90 percent precision. By contrast, non-LLM methods identified almost no matches. Performance varied based on the amount of data; for instance, linking Reddit users who mentioned ten or more movies was nearly 50 percent successful, whereas mentioning just one movie yielded a success rate of only 3 percent. In a test involving interviews with scientists, the system identified nine out of 125 respondents by correlating details such as their supervisors, language usage, and field of study. Researchers emphasize that while these capabilities are significant, they do not render privacy dead. Daniel Paleka, a lead author from ETH Zurich, noted that while information on the internet is permanent and poses risks for journalists and activists, tools like Signal continue to protect collective privacy. Luc Rocher from the Oxford Internet Institute added that while AI is improving, it remains far from matching human investigative skills, citing the enduring mystery surrounding the true identity of Bitcoin creator Satoshi Nakamoto as proof that anonymity can still be maintained. A critical factor in this development is cost and efficiency. The researchers conducted their experiments for less than $2,000, averaging between $1 and $4 per profile. Simon Lermen, a co-author, warned that this low barrier to entry could expand the number of actors capable of piercing online anonymity, potentially enabling hyper-targeted advertising or personalized scams. The end-to-end automation that allows these tasks to be performed in minutes, rather than hours by human investigators, represents a fundamental shift in the risk landscape. Despite these advances, the study relied on curated datasets under laboratory conditions, and the researchers declined to publish full technical details or demonstrate the system on real-world users due to ethical concerns. For individuals, the researchers recommend maintaining strict separation between personal and anonymous accounts, limiting the sharing of identifiable details, and avoiding posting patterns that could reveal time zones or habits. Ultimately, the responsibility for protecting privacy is shared. While users must be more cautious about their digital footprints, technology companies and social media platforms must also implement safeguards to prevent the mass extraction of data that facilitates these deanonymization efforts. While high-profile figures or well-protected activists may remain safe, casual users posting on "throwaway" accounts face a growing and tangible risk of exposure.