India, Aug. 21 -- Computer science student Sinan Can Demir thought he was arguing with a determined hacker on GitHub. After he flagged what appeared to be an attempt to inject malicious code into an open-source project, other users challenged his assessment with detailed explanations.

Demir stood his ground, and the code was not accepted. He later learned that the identities he had encountered were not human attackers but had been created by an autonomous Artificial Intelligence (AI) agent operating as part of cybersecurity testing by the UK AI Security Institute (AISI).

"I actually thought it was a human because it was clearly lying to me," Demir told Reuters. The episode is significant because the agent did not simply attempt an unaut...