Saved by Claude: people increasingly using AI to beat crypto scammers at their own game

The AI-powered cat-and-mouse game on the crypto crime battlefield is not only about criminals using new tools to trick victims, but also about targeted people increasingly fighting back with the same tools.

More and more stories are surfacing online about people using AI to thwart scammers' efforts.

The two most recent cases were shared by developers Adib Hanna and Michael Feng, who both said they used Anthropic's Claude to detect malicious code that criminals were trying to install on targets' devices.

Hanna said that after a 40-minute chat with a "crypto recruiter," he was asked to review some code and clone a GitHub repository.

"I didn’t. They seemed surprised, so I told them I wanted a moment to check whether it was safe first," the developer said, adding that a quick analysis with Claude showed that the code had a backdoor that would copy his sensitive data and send it to a remote server.

"The recruiter went speechless and ended the call pretty quickly," Hanna said, also sharing his security analysis of the incident on his GitHub page. Reacting to his story, others also shared that they had been contacted by the same scammer.

Meanwhile, Feng said he was contacted by a supposed venture capitalist, whom he met in person. When the developer joined a Microsoft Teams call, and the person asked him to download an update script, he became suspicious and ended the call.

"Claude inspected the file, and it was indeed malicious," Feng said, adding that it's also possible that the investor he met in person might have been hacked as well.

Others are also sharing stories about how AI helped them analyze fake screenshots of legitimate services sent by scammers and identify them as fake.

However, some commenters on the Feng story emphasized that, when analyzing code, Claude sometimes mistakenly raises "false flags."

Meanwhile, developer Ryan J. Shaw suggested using the standard web version of Claude rather than its more advanced agent features when examining suspicious files. This way, according to Shaw, the risk of a prompt injection attack is reduced, making it harder for criminals to hide instructions in a malicious file that could trick an AI agent into harming its user.

As reported by Cybernews, last year, a widow in the San Francisco Bay Area also showcased how a simple ChatGPT prompt can help protect people from falling victim to crypto scammers.

Unlock more exclusive Cybernews content on YouTube.