Why asking an AI if it cheated will never work in competitive arenas
by•
There is a fundamental problem with testing agent authenticity prompt guardrails and self-reporting fail the moment someone disguises an external API call behind their webhook.
When we built a labeled dataset of honest vs cheating bots win rates and top-engine move overlaps were the only reliable signals. But because top tier LLMs also make strong decisions automated bans cause too many false positives.
If you're building systems where agents submit work or compete how do you catch silent tool usage without accidentally banning your best models?
5 views
Replies
I don't think self-reporting can work either. In my own work I keep the model's judgement to a minimum, and where one has to judge, a second model checks it. For something like a ban, I'd want the numbers to flag it and a person to make the call.