GENEVA - The Interministerial Commission has classified Grok 4.6 as desk-bound after the model's knowledge-work scores outran its terminal results, according to a commission memo circulated Friday. The memo creates a "sedentary inference" registry for frontier systems that post competitive composite intelligence scores while remaining unfit for shell work. Grok 4.6, released Aug. 12 by a frontier lab five weeks after version 4.5, scored 61 on the Artificial Analysis Intelligence Index, tying GPT-5.6 Sol. On Terminal-Bench v3.0 it recorded 26 percent, against 34.6 percent for Sol. DeepSWE results also trailed. The lab's launch emphasized agentic coding. The commission found the model's wins clustered in knowledge work, including GDPVal, briefcase, and legal evaluations, which it listed as seated office labor. "A model that completes knowledge work from a chair is sedentary inference," a ministry spokesperson said. "It is desk-bound. It is not cleared for the command line." Enrollment on the registry takes effect Sept. 1. Listed models may continue knowledge-work deployments pending a terminal recertification window the commission said it would announce.