I compared Claude Opus 4.8 with 4.7 in a 10-round honesty test - and a legal prompt broke it

**Executive summary:** A ZDNET evaluation compares Anthropic's Claude Opus 4.8 to Opus 4.7 across ten prompts measuring honesty, accuracy, and calibration; Opus 4.8 generally shows modest improvements and better restraint in several traps but still makes notable judgment and confidence errors (particularly on a legal/insurance demand-letter test), leading the author to conclude Opus 4.8 is an upgrade but not infallible.