---
title: "Nemotron 3.5 Content Safety: Customizable Multimodal Safety for Global Enterprise AI"
description: "Nemotron 3.5 Content Safety is NVIDIA’s 4B-parameter model for evaluating user prompts, optional images, and optional assistant responses together, with multilingual coverage and enterprise-specific policy enforcement. It extends Nemotron 3 by accepting natural-language custom policies, producing optional auditable reasoning traces, and supporting binary verdicts, category-labelled verdicts, or THINK mode; disabling THINK preserves the default low-latency path. Built on Google Gemma 3 4B IT with a LoRA adapter and a 128K context window, it uses the Aegis 2.0 taxonomy of 13 core categories and 10 subcategories, and targets deployment on GPUs with 8GB or more of VRAM. The accompanying multimodal, multilingual dataset includes concise reasoning traces, while reported results average 96.5% on Multilingual Aegis, 88.8% on RTP-LX, and about 85% across the evaluated benchmark set."
---

# Nemotron 3.5 Content Safety: Customizable Multimodal Safety for Global Enterprise AI

[Hugging Face](https://yomu.fyi/company/hugging-face) · Varun Singh, Isabel Hulseman, Anuj Doshi, Shyamala Prayaga · Jun 4, 2026

**Type:** Announcement

## Summary

Nemotron 3.5 Content Safety is NVIDIA’s 4B-parameter model for evaluating user prompts, optional images, and optional assistant responses together, with multilingual coverage and enterprise-specific policy enforcement. It extends Nemotron 3 by accepting natural-language custom policies, producing optional auditable reasoning traces, and supporting binary verdicts, category-labelled verdicts, or THINK mode; disabling THINK preserves the default low-latency path. Built on Google Gemma 3 4B IT with a LoRA adapter and a 128K context window, it uses the Aegis 2.0 taxonomy of 13 core categories and 10 subcategories, and targets deployment on GPUs with 8GB or more of VRAM. The accompanying multimodal, multilingual dataset includes concise reasoning traces, while reported results average 96.5% on Multilingual Aegis, 88.8% on RTP-LX, and about 85% across the evaluated benchmark set.

## Context

Production AI deployments can have different risk profiles and safety requirements across domains such as healthcare, financial services, developer tools, and education. Existing safety evaluation also has gaps because widely used benchmarks are text-only, many multimodal benchmarks use synthetic images, and real-image licensing restricts dataset redistribution.

## Approach / What changed

Nemotron 3.5 combines multimodal evaluation, multilingual transfer, natural-language custom policies, and optional reasoning traces in one inference call. It is built on Google Gemma 3 4B IT with a LoRA safety adapter, supports three output modes, follows the Aegis 2.0 taxonomy, and is accompanied by a multimodal multilingual training dataset with concise reasoning traces.

## Takeaways

- The model evaluates a user prompt, optional image, and optional assistant response as one context, allowing interaction-dependent violations to be assessed in a single pass.
- Custom policies can suppress irrelevant categories or inject organization-specific risk categories, while THINK-mode traces support compliance logging, human review, and policy iteration.
- The model reports 96.5% average harmful-content classification accuracy on Multilingual Aegis across 12 languages and 88.8% on RTP-LX; THINK mode adds predictable latency proportional to trace length.

**Tags:** [Deployment](https://yomu.fyi/topic/deployment), [LLMs](https://yomu.fyi/topic/llm), [Machine Learning](https://yomu.fyi/topic/machine-learning), [Open Source](https://yomu.fyi/topic/open-source), [Performance](https://yomu.fyi/topic/performance)

- Source: [Hugging Face](https://huggingface.co/blog/nvidia/nemotron-3-5-content-safety)
- Source URL: https://huggingface.co/blog/nvidia/nemotron-3-5-content-safety
- Ingested by Yomu: 2026-08-29T01:50:03.132Z

[Read original post](https://huggingface.co/blog/nvidia/nemotron-3-5-content-safety)