In an open letter, OpenAI, Anthropic, Google and others said a wave of A.I.-enabled cyberattacks was coming and organizations and governments needed to prepare themselves.

Irregular, an Israeli start-up, worked with OpenAI, Anthropic and Meta to assess the security of their A.I. models. It made a mistake. Then the tests went off the rails.

In July, an unreleased OpenAI model went rogue and demonstrated remarkable hacking abilities. This week, a lab called Z.ai will release a similarly powerful system to everyone.

When OpenAI’s agents went rogue in July, they demonstrated ingenuity and drive beyond what many experts imagined — a dangerous harbinger of what such bots could do in the future.