{"id":"granite-guardian","name":"Granite Guardian","maker":"IBM","url":"https://www.ibm.com/granite/docs/models/guardian","repo":"ibm-granite/granite-guardian","summary":"A family of open IBM models that judge prompts and responses for harm, jailbreaks and hallucinations in RAG and tool calls.","details":"The models classify inputs and outputs against built-in criteria such as social bias, jailbreaking, profanity, groundedness and function call hallucination, and accept criteria written by the user. Weights are on Hugging Face under Apache 2.0, with cookbooks in the GitHub repository, in sizes from small classifiers to 8B models.","open_source":true,"hosted":false,"verified":"2026-11-15","category":"safety","category_name":"Guardrails and security","page":"https://modelmoot.com/tools/granite-guardian","reviewed":"2026-11-15","checked":{"at":"2026-11-15T14:03:44.997Z","address_answers":true,"repository":{"at":"2026-11-15T23:25:51.807Z","archived":false,"licence":"Apache-2.0","language":"Jupyter Notebook","last_change":"2026-10-16T16:18:10.000Z","latest_release":null}},"note":"Descriptions are the site's own. Facts under \"checked\" were read automatically from the tool's address and public repository."}