Artificial intelligence is opening new possibilities for businesses and productivity. But it is also creating new risks for privacy. A recent study suggests that large language models, the technology behind tools like ChatGPT, can make it significantly easier for hackers to identify anonymous users on social media.

Researchers Simon Lermen and Daniel Paleka tested how effectively AI systems could match anonymous accounts with real identities across different platforms. In many scenarios, the models were able to connect profiles by analyzing small pieces of publicly shared information and searching for similar patterns elsewhere online. What previously required extensive manual investigation can now be done quickly and at low cost using AI.

For example, a hypothetical user posting anonymously about struggling in school and walking their dog “Biscuit” through Dolores Park could unintentionally reveal enough clues for an AI system to identify them. By scanning other platforms for similar references, the model could match the anonymous handle with a real identity with high confidence. While the example is fictional, the technique reflects how AI can combine scattered online signals to build a detailed profile of a person.

This capability raises concerns for both privacy and cybersecurity. Researchers warn that governments could potentially use such tools to monitor activists or dissidents posting anonymously, while cybercriminals could leverage the same techniques to create highly personalized phishing scams. AI-powered surveillance is becoming a rapidly evolving field, allowing systems to analyze and synthesize vast amounts of publicly available information that would be difficult for humans to process manually.

Even today, publicly available personal data can already be used in targeted scams such as spear phishing, where attackers impersonate trusted contacts to trick victims into clicking malicious links. With AI lowering the technical barriers, attackers may only need access to publicly available language models and an internet connection to launch sophisticated attacks.

Experts also warn about the broader implications. Peter Bentley, professor of computer science at University College London, notes that commercial tools designed specifically to de-anonymize users could emerge in the future. However, he also cautions that AI systems are not perfect. Large language models can make incorrect associations, potentially linking the wrong identities and leading to false accusations.

Another concern is the growing ability of AI to combine information from different public sources beyond social media. Marc Juárez, a cybersecurity researcher at the University of Edinburgh, points out that hospital data, admission statistics, or other public datasets could potentially be cross-referenced by AI systems in ways that compromise anonymity.

Despite these risks, AI is not a guaranteed method for revealing identities. The process still depends on the amount and consistency of information people share across platforms. If there are too many potential matches or insufficient clues, the models cannot reliably identify individuals.

Still, the research highlights a growing challenge for both organizations and individuals. As AI systems become more powerful at connecting data points, traditional approaches to anonymizing information may no longer be sufficient.

The researchers suggest several practical steps for platforms to reduce the risk. These include limiting how quickly users can download large amounts of data, detecting automated scraping activity, and restricting bulk exports of user information. At the same time, individuals may need to become more cautious about the personal details they share online.

In an AI-driven world, even small pieces of information can combine into a much larger digital footprint. Protecting anonymity is becoming less about hiding data entirely and more about understanding how seemingly harmless details can connect across the internet.

Source