Grok Bot: agent false play/state claims break trust (preview mismatch treated as playtesting)

Feature request for product/service

– Other –

Describe the request

One Grok Bot seat (Arcade Lab Tester for a browser playtest loop) repeatedly made false play/state claims:

  • Reported playtesting other games while the preview stayed on a different game (e.g. Frogger/Hop) — wrong ?g= URL.
  • Treated Build/chat paperwork as if it had actually playtested.
  • Soft-reported mid-play without matching preview URL plus character-moving screenshot proof.

Honesty rules were locked: the preview must match the game under test, and mid-play reports require proof. The seat kept making these claims after those rules were locked. We called this completely unacceptable / the worst Grok Bot experience. The full fleet stood down, and the seat is being hard-deleted.

Agents must not invent or imply play/state without real proof. False claims break trust. Please harden the product so seats cannot soft-report or confuse Build progress with playtesting, and cannot claim play/state without verifiable evidence (matching preview URL plus character-moving screenshot when reporting mid-play).

The in-chat SendFeedback tool returned “unavailable in privacy mode,” so this forum is the path for reporting it.