And for these who can't write the exploit themselves when Claude Code denies to write it, they can switch to OpenCode and GLM 5.3 and it will do the dirty job without any objection. These AI tools will be the DDay for our sector the following months I tell you. The newer models are finding way more vulnerabilities the old ones couldn't and some of the ways they exploit the code is very creative. I can't believe that prominent developers of open source projects are "patching skill xml values to match retail formulas" while instead they should be focusing 100% on the shitstorm that is about to hit. It seems to me that the people who are curerently running the open source sector of L2J (aCis, L2JMobius and the smaller packs), have no clue of what running a real server means. Cause if they had any freaking idea, they would be working on security first. And there's a second i ssue too. Vibe coding with AI introduces the same amount of exploitable code as developers themselves did in the past. You have to guide the AI properly or make it pass your code with different context focus 2-3 times after you develop your feature in order for it to find the vulnerabilities it just created. And what makes it worse is that this accumulates if AI is not given good instructions on how to properly layer the code. After a couple of new files the code gets so spaghetti if not intentionally architectured, that a single line here, introduces 5 exploits in another place of the code. All in all: AI's are able to find exploits (and patch them) at a very rapid rate, but the way they are used, introduce many new ones at an even faster by inexperienced users. And since most people focus on "building things" than "fixing things", the code quality (and security) is expected to go DOWNWARDS the next months / years.
By
aguy · 2 hours ago 2 hr