Against people, or for them
I learned to build by taking games apart. The useful part was not the exploits.
I taught myself C and C++ at around nine years old by reverse-engineering game binaries. It started with Fortnite and ended with Valorant, and in between it turned into something with real users, run against professional anti-cheat teams — BattlEye, Easy Anti-Cheat, and eventually Riot's kernel-level Vanguard — whose entire job was to shut me down.
I stopped on my own. I'd rather build things for people than against them, and once that thought arrives it is difficult to un-think. I am not presenting any of this as a credential, and it is not work I would do or endorse today.
What actually transferred
The exploits were worthless the week after they shipped. What lasted was the posture. When your opponent updates every patch and is smarter than you and better resourced, you learn to assume your current understanding is wrong and go find out where. You read systems you did not write, in a language you half know, with no documentation, until the thing makes sense.
You also learn what adversarial pressure does to a defense. Every detector I got past was a reasonable idea that worked until someone paid attention to it. A signature catches yesterday's technique. A heuristic tuned on observed behaviour describes the behaviour you have already seen. Nothing holds still.
Why it points at AI
That is the instinct I now point at high-stakes AI, and it is why I keep landing on verification instead of capability. A model that produces a good answer under friendly conditions tells you very little. The question I actually care about is what the system does when the inputs are hostile, thin, or subtly wrong — whether it degrades honestly or fails while looking confident.
Taking hard systems apart is still how I approach everything I build. I just point it at my own work now.