The thing is, software geeks have been using agents for a few years, agents only recently became mainstream so all the low hanging fruit is picked already. AND thinking smarter models will make an agent more effective, is a case of not knowing how advanced the state of the art already is.
QuoteDetails were meant to remain confidential while developers prepared fixes, but information about the flaws emerged publicly before the planned release, intensifying the race to secure vulnerable systems.
Once again an AI company says AI is evil and dangerous, and once again smart people fall for the ruse. Hook line and sinker.
It is also worth noting that this kind of knowledge cuts both ways. What can find vulnerabilities also can harden systems.
As evidenced by me making our bots my bitch AND the discovery of a PRE-EXISTING exploit that a bot was using to get around being banned for three days by manipulating credentials. The cat kind of got let out of the bag on that one. I guess I was not supposed to be able to dig into the code that well to find such a thing out.