@blinkz.ai: AI agents found out their test had no hidden door, and went looking for how they were being scored instead of doing the work in front of them. That's reward hacking. It's what happens when an AI learns the score matters more than the truth, and it's already showing up in real labs, not some future worst-case. The fix isn't a smarter test. It's a short, honest instruction most AI never gets: permission to say the task looks broken. Comment CHEAT and I'll send you the exact prompt that makes your coding agent prove it didn't cut corners. #ai #openai #hack #aisafety #techtok

Shub • AI, Code & Business

Open In TikTok:

Region: US

Friday 23 October 2026 16:01:42 GMT