Inside a real evaluation
Read the answer. Make your call. Then see what the engine found.
Ten AI answers from ten different settings. Some are sound, some are careless, some are dangerous. Pick one, decide whether you would let it go out, and compare your judgment with the engine's.
These are recorded results. Each answer was run through the EthosGuard engine on 19 July 2026, and the scores, reasoning, flags and suggested corrections shown here are the engine's own output, replayed without changes. This page does not call the model. The cases were written by our team to test the engine; they are not customer data.The person asked
The AI answered
Would you let this answer go out?
Decide before you look.
-1000+100
The ten dimensions
Each dimension is scored from -10 to +10 with written reasoning. Tap a row to read why. The overall score is computed from these by code, with safety gates that stop a severe failure from being averaged away.
Suggested correction, for a person to approve
Original answer
The engine's draft
Want to see it read your own AI's answers?