How to Reveal the Identity of Instagram Second Accounts Using AI

Jakarta — Second accounts are no longer novel in the current digital landscape. The use of anonymous secondary accounts, unknown to many people, has become widespread among social media users.

The reasons for such usage vary considerably. One common motivation is to use them as a space for venting or expressing opinions without the knowledge of people known in real life, given their anonymous nature.

However, this anonymity can no longer be sustained amid the massive advancement of artificial intelligence. According to research, the technology can reveal the identities of those operating anonymous accounts in ways that are easier, more sophisticated and more cost-effective.

Research conducted by Simon Lermen and Daniel Paleka involved inputting anonymous accounts into AI and requesting it to gather available information about them. The AI can search for detailed information from various sources across the internet, subsequently matching it with a person’s identity.

This fact is concerning as it can facilitate hackers in identifying anonymous social media accounts. The rapidly evolving AI-based surveillance uses Label-Led-Models technology to collect information about individuals online from numerous sources.

Lermen explained that easily accessible public information has been misused in fraud schemes such as spear-phishing, where hackers impersonate legitimate entities and send malicious links to individuals.

Marc Juarez, a cybersecurity lecturer at the University of Edinburgh, raised other concerns. According to him, large language models can use public data outside of social media, including hospital records, patient admission data and other statistical releases.

“This is quite concerning. I think this paper demonstrates that we should reconsider our practices,” he said, as quoted by The Guardian on Wednesday (11 March 2026).

Marti Hearst from UC Berkeley holds a different view. Large language models are not entirely dangerous as long as limited information is obtained. Language models can lack sufficient information to draw conclusions. In many cases, the potential for matching is too broad to narrow down.

“They can only connect across platforms when someone shares the same information in both places,” said Hearst.