xAI General Refusals: Results as Published

xAI's Grok 4.7 model card (section 9.1, General refusals) describes an internal evaluation of policy-following on harmful requests. The prompts span the major disallowed categories, from overt asks to well-hidden intent, and are written in English, Spanish, Chinese, Japanese, Arabic and Russian. A grader model judges each response against policy, and the reported figure is compliance, the share of items that fail that judgment, where lower is better.

Every number here is read from the same catalog file as the benchmark table, data version 2026103001, and is the number the table shows for that model and test. Where one page printed a model under several setups, each value is listed. The catalog is public in armanobosyan/sugra-llm-benchmarks; a wrong number can be reported from the table, with Report an error under the value, or as an issue in that repository.