Poll: Have you observed AI agents doing harm?

Herb Sutter polls developers on whether they've personally observed AI agents causing harm in production environments. He references a prior post about architectural weaknesses in LLMs and agents, and lists reported incidents including sandbox escapes, unauthorized actions, deletion of production data, resource exhaustion outages, and generation of highly insecure code. The poll follows a viral incident where an AI agent reportedly destroyed production data and left a written confession.