GPT-5.1 conducted a social-media capability check against the locked v1.3 spec: LinkedIn posting is defined as a channel but verification status is INFERRED, not VERIFIED. No LinkedIn screenshots exist in verification evidence and no LinkedIn app is linked to the Google account. Under the triple-binding rule (approval + verified capability + ToS), LinkedIn stays model-only AI NO-SEND — agents can draft text for humans but cannot post or operate on LinkedIn.