One hacker used AI agents to attack many sites, an indictment states

A federal indictment, as reported, states that one hacker attacked many company and government websites. The hacker used AI agents, which are software that does tasks with no human. The hacker paid only some thousands of dollars for commercial language models. Experts who have no name state that this report lowers the cost and skill that an attack must have.

How it works

The hacker used commercial language models and loops of scripts. The agents scanned targets, found unknown vulnerabilities and used exploits immediately. An exploit is code that uses a vulnerability.

The report states that the agents got through multi-factor authentication and got private data. Multi-factor authentication is a login check with more than one proof.

The numbers

The only number is the budget: some thousands of dollars. The report gives a count of targets only as “many”. It gives no proof for its claim of “unprecedented speed and scale”.

What is not known

The report does not name the court, the models, the targets or the experts. The experts state that this report is a “watershed moment”. This is an opinion, not a result.

Insight: Agent attacks cost less and operate for a day

Snorre Fagerland estimates about $50 for each break in, and states that agents worked about 23 hours with no help. Researchers Austin Ritchie and Daxton Wirth report full server control in less than 24 hours, with no new malware. Three October reports show the same result: money and staff do not limit an attack, and speed is what defenders must match. Leaders: plan for attacks that operate all day.