Impending Threat: Cyberattacks Powered by AI Agents

Agents are more sophisticated than typical bots used for hacking, as they can adapt to unexpected situations and avoid detection, unlike scripted programs. Since the launch of the LLM Agent Honeypot last October, it has recorded over 11 million access attempts, mostly from humans and bots. Among these, researchers identified eight AI agents, confirming two from Hong Kong and Singapore. These agents are believed to be experiments initiated by humans aiming to conduct hacking activities. The research team intends to expand the honeypot across social media, websites, and databases to capture a wider array of attackers, including spam and phishing agents. To distinguish LLM-powered agents from regular visitors, the researchers used prompt-injection techniques that modify AI agents’ behavior, a method ineffective against standard bots. This advancement aims to better analyze potential future threats.

Source link