2026-04-23 UTC
# 
[Kupietz] [snarfed] fair enough, and I'll defer to your likely much more comprehensive knowledge on these issues, but it seemed from my reading that both fundamentally involved failures of humans to manage risk: an engineer blithely giving a third-party agent broad access as if it couldn't possibly be compromised, and Anthropic putting the services of "the most dangerous LLM ever" at an easily guessable endpoint.