theaicatalog
Hugging Face ยท Open source

IBM Granite releases LLMON, a markup language meant to block prompt injections

IBM Granite has released two experimental preview models, granite-4.1-3b-LLMON-preview and granite-4.1-8b-LLMON-preview, which are trained to read and write LLMON, a markup language that marks instructions, data, and their links for the model. IBM says the models mostly ignore indirect prompt injections hidden in user data, and on the OPI benchmark they avoid 47% more attacks than the best frontier model tested. The company says general performance stays on par with the Granite 4.1 Instruct models. The authors describe the preview as ready for exploration and evaluation but not for production.