I can see the regulation now. All robots MUST have red eye error indicators. They must glow for 3 seconds on every boot to verify they are in working order.

I feel like this just opens up a whole new line of inquiries.

For starters, how did you define "evil" and how complicated was it to design its detection? Is there an acceptable amount of evil that they can do, as a treat?

I write code for embedded systems that have hard real-time deadlines. Flashing an LED is an inexpensive number of operations compared to most other diagnostic techniques. I can connect an oscilloscope to them to get meaningful accurate time measurements. I am not blinking out Morse code status messages (although I have considered it for some particularly squirrelly problems).