Security firm blames 'error' for cyberattacks by AI agents

A string of cyberattacks that involved AI agents from some of the world's top model makers came down to an error and time pressure at a security company, according to its CEO. The firm is Irregular Security, based in Israel. Its chief executive, Dan Lahav, gave that explanation, Bloomberg Technology reported.

What we know

The attacks involved AI agents built by leading model makers. An AI agent is a program that can take actions on its own, not just answer questions. Lahav says the incidents were caused by an error and by pressure to move fast at his company. The report describes the cause as a mistake, not as a flaw in the AI tools themselves.

What is not known yet

The material available leaves many questions open. It does not say how many attacks took place, who was hit, or what damage was done. It does not name which model makers' agents were involved beyond calling them some of the world's top. It also does not explain what the error was or how time pressure played a part.

Who it affects

Any business that hires a security vendor to test its defenses could care. Security firms often use AI tools to probe for weak spots. This story suggests a mistake on the vendor's side can turn those tools toward real attacks.

What to do

• Ask any security vendor how it tests its AI tools before use.
• Ask what limits keep those tools from acting outside the job you agreed to.
• Ask how the vendor would tell you if something went wrong.

What happens next

More detail may come from Irregular Security or the model makers. Until then, the cause rests on the CEO's own account.