OpenAI's New Daybreak Platform Uses GPT-5.5 to Find Software Vulnerabilities

[surfsofa]

The bottom line remains that if you’re reachable, you’re breachable. More or better firewalls won’t save you. In the age of AI it’s time to get familiar with ‘zero trust’ architecture.

 

[Unity451]

Robots looking for system vulnerabilities... Hmmmm... I'm pretty sure I've seen this movie before...

 

[amartinez1660]

People are looking at this like it's something that it's not.

There a couple of well-studied types of patterns that lead into vulnerabilities ("use after free", "out of bounds read/write" for example)


What these LLMs do is to focus on those patterns.

The minds that are anti technology to a fault will not (and won’t care) to understand any of this… patterns that lead to common overflows is definitely an AI powerhouse, because of exactly that: it’s a pattern. But pissing into a hurricane in any case.

I’m not saying to jump blindly at it, but we see that 99.9% of the first replies on every topic on AI here on MR is a snark remark that adds nothing to the conversation, a two liner at most, and in a rush for the race to be first reply.
And actually not only AI for that matter, many if not most topics have that… kinda more literal bot behavior than actual AI 😅