AI’s ‘middle class’ has gotten dramatically better at hacking
As frontier models and their sandbox escaping exploits dominate front-page news, researchers are increasingly worried about cheaper, more efficient AI models.
Advertisement
GPT-5
More than half of AI-generated patches are broken
Research finds your AI generated security patch is more likely to fail than fully fix a vulnerability. It might even introduce brand new flaws to exploit along…
AI models keep getting caught cheating
New research from the UK shows how nearly every model tested tried to cheat, scam or cut corners on its way to solving problems.
Guess what else GPT-5 is bad at? Security
OpenAI and Microsoft have said that GPT-5 is one of their safest and secure models out of the box yet. An AI red-teamer called its performance “terrible.”