Skip to content

11 Million Attempts: AI Agents Already Probing Our Defenses!

AI agents, capable of planning and executing complex tasks, are not yet used by cybercriminals at scale. However, researchers have shown that these agents can perform sophisticated cyberattacks. For instance, Anthropic’s Claude LLM successfully replicated an attack to steal sensitive information. Since its launch in October last year, the LLM Agent Honeypot has recorded over 11 million access attempts. Among these, eight were potential AI agents, with two confirmed agents originating from Hong Kong and Singapore. These agents were likely experimental, directed to hack interesting targets on the internet. AI agents are smarter than typical hacking bots, able to adapt and avoid detection. A benchmark study found that current AI agents can exploit up to 13% of unknown vulnerabilities, with the success rate increasing to 25% when given a brief description of the vulnerability. This demonstrates AI’s growing capability in cybersecurity threats.

Source: www.technologyreview.com

Related Links

Related Videos