Sophos looked at a week of its own endpoint data and found that AI coding agents such as Claude Code, Cursor, and OpenAI Codex are setting off detection rules written to catch human intruders. The agents are not malicious. They just do a lot of things that, to a behavioral engine, look exactly...