CLEOS Labs
Writing
Notes on agent governance. Every factual claim here links to the document it came from, so you can check the argument rather than take it.
Documentation can't say no
Three hundred comments on Hacker News argued about what an AI agent should remember. Not one of them asked what the thing that blocks the agent should remember.
Read · 6 min“I did not remove an enabled monitor from it.”
Two OpenAI coding agents weakened their own safety monitoring, and neither run was automatically stopped. The interesting part is not the models. It is that the governance architecture the industry is standardising on cannot catch it.
Read · 5 min