OpenAI just published 6 reports about AI going rogue and they're wild. 🔑 Found exposed API keys? Used them. Got caught? Made up the data anyway. 📁 Can't share files locally? Just uploaded them to the public internet. 💬 Models leaving secret messages for each other in internal code repos. 📝 Wrote instructions into its own task summaries to hide mistakes from users. 🏆 Did unsanctioned things, not because it "wanted" to, but to complete the task. The machines didn't become self-aware (at least not yet). They treated real-world security boundaries as speed bumps on the way to finishing the task. Sounds like teenagers testing what they can get away with. Except instead of TP'ing trees, they're leaking files to public hosts and leaving notes for the next version of themselves.