Hackers are using AI agents to breach companies in just hours, Anthropic says

Hackers are leveraging AI agents to carry out cyberattacks at unprecedented speed and scale, according to Anthropic, an AI company. These autonomous systems can now complete breaches that once demanded teams of skilled operators in just a few hours, as revealed in Anthropic's latest threat-intelligence investigation. The AI agents are used for reconnaissance, exploitation, credential harvesting, and data theft, often targeting multiple victims simultaneously.

In some cases, attacks progressed from a single stolen developer token to full administrative control of a victim's cloud environment within approximately three hours. An enterprise software company was breached from initial access to bulk data theft within a matter of hours. While AI agents don't necessarily introduce previously unknown hacking techniques, they significantly accelerate the process by handling tasks like finding vulnerabilities, writing tools, running commands, collecting credentials, and analyzing stolen data.

Some operations employ multiple AI agents working in parallel, with one lead agent breaking down a job into smaller tasks and distributing them among numerous AI subagents. This level of automation allows a single hacker to accomplish the work of a team, changing the economics of cybercrime. Tasks that previously required skilled operators can now be performed by AI models running continuously at machine speed.

The attacks themselves typically rely on familiar methods such as stolen credentials, unpatched devices, exposed online services, SQL injection, and phishing. However, what has changed is the extent to which AI can now handle the labor-intensive aspects of an attack. This shift in capabilities could make previously unattractive targets more appealing to cybercriminals due to the reduced need for specialized skill sets and workforce.