Among Mythos’ cybersecurity capabilities, researchers have found that GPT-5.5 is also good.

Is it just “fear trading”?

The new GPT-5.5 results show that, when it comes to cybersecurity risk, the Mythos Preview may not have been “a one-off success” but “the result of a long-term evolution of independence, reasoning, and documentation,” AISI writes.

In a recent interview with the Core Memory podcastOpenAI CEO Sam Altman criticized what he called “fear mongering” in promoting limited releases of some AI models. While he said that he “certainly believes that Mythos is a good example of cybersecurity,” he added that “it’s clear that it’s not surprising that a business would say, ‘We’ve built a bomb, we’re about to drop it on your head.’

“There’s going to be a lot of talk about the kinds of things that are too dangerous to come out,” Altman continued. “There will also be more dangerous examples that need to be released in different ways.”

In February, OpenAI released its Trusted Access for Cyber ​​test programallowing security researchers and businesses verify their IDs and register their interest in studying OpenAI’s examples of “legal defense work.” Last month, OpenAI said it was using a trusted access list manage the minimal installation of GPT-5.4-Cyber.

On Thursday, OpenAI CEO Sam Altman he said on social networks that the first release of GPT-5.5-Cyber ​​will also be “for cyber defenders in the next few days.”