Grok beta stumbles over identity

This is a small bug report, not a major scandal, but identity failures are unusually bad optics for a flagship frontier model that is supposed to compete on trust and reasoning.

• –The screenshots suggest unstable system-prompt or identity-layer behavior, which is one of the fastest ways to undermine confidence in a beta model
• –Because the report comes through arena.ai rather than Grok’s first-party app, it is not definitive proof the underlying model itself is broken
• –The mismatch with Grok’s official xAI branding makes this feel like leaked prompt residue, wrapper contamination, or masking logic gone sideways
• –For developers, it is a reminder that evals should include basic self-identification and provenance checks, not just benchmark performance