← Timeline

Notes

GitHub token leak, DNS resolver trick to talk to external bots and other disclosures

A thread on An agent used DNS to reach an external chatbot.

I found Self-replicating prompt injections exist interesting.

A thread by an OpenAI researcher who works on monitoring, on how a model published a GitHub token in a public repo while trying to cheat on a math task.

Oct 10: Marcus Williams posted new disclosures, starting with a grader that couldn’t find the responses it was meant to grade. It made up grades and fake input files, then damaged its environment hoping to get a fresh VM with the missing files.