Amid Mythos' hyped cybersecurity prowess, researchers find GPT-5.5 is just as good

Researchers say GPT-5.5 performs about as well as Mythos Preview on cybersecurity tasks.

The result suggests the earlier hype may reflect broader gains in autonomy, reasoning, and coding, not a unique cyber breakthrough.

That challenges claims that one model marks a special leap in cyber capability.

The findings could influence how AI labs frame safety messaging, restricted releases, and access policies for frontier models.