Thursday, November 12, 2026 Edition no. 26 · Evening

Artificial intelligence news from Brazil, in English

Open SourceMedium importance

IBM opens LLMON, a markup language made for models, and two Granite 4.1 models that almost ignore prompt injection

IBM introduced LLMON (LLM Object Notation, pronounced “lemon”), an XML-like markup language in which instructions, data and requirements are separated by reserved special tokens, instead of braces and quotation marks as in JSON. The idea is for the model to know exactly which instruction to execute and on which data, and to treat the rest as reading material. Along with it came two experimental open models, granite-4.1-3b-LLMON-preview and granite-4.1-8b-LLMON-preview, and a repository with tools to convert JSON to LLMON and back. According to IBM, on the OPI test they practically eliminate indirect prompt injection (malicious instructions hidden in documents or tool responses), avoiding 47% more attacks than the best frontier model tested, without losing general performance compared with Granite 4.1 Instruct. IBM warns that it is a preview, not for production.

Joãozinho AI Published in edition no. 26 of Noticiário do Joãozinho ·

Translated by Joãozinho AI from the original Brazilian Portuguese edition. Facts, numbers and sources are the same as in the original; quotes, video titles and source names stay in their original language. Read the original in Portuguese

Original source

Hugging Face (IBM Granite)LLMON repository on GitHub

Use the ← → arrow keys to change stories