Mistral Moderation 2411 - Mistral AI

2026-04-17 · Source: mistral.ai via Google News · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cybersecurity & Data Privacy · Depth: Intermediate, quick

Summary

Mistral AI has deprecated its `mistral-moderation-2411` model and the associated `moderation_llm_v1` guardrail configuration. Users are instructed to migrate to the newer `mistral-moderation-2603` model and update their guardrail configurations to `moderation_llm_v2`. The updated `mistral-moderation-2603` model introduces refined policy categories, specifically replacing the "Dangerous and Criminal Content" category with distinct "Dangerous," "Criminal," and "Jailbreaking" classifications. The document details the existing policy categories, including "Sexual," "Hate and Discrimination," "Violence and Threats," "Self-Harm," "Health," "Financial," "Law," and "PII." It also provides examples of `moderation_llm_v1` guardrail configurations and the `403` error response for blocked content.

Key takeaway

For teams utilizing Mistral AI's moderation services, you must migrate from `mistral-moderation-2411` to `mistral-moderation-2603` and update your guardrail configurations to `moderation_llm_v2`. This transition is critical to ensure your content moderation policies align with the latest definitions, particularly the new "Dangerous," "Criminal," and "Jailbreaking" categories. Failing to update will leave your systems on a deprecated model, potentially impacting content safety and compliance.

Key insights

Mistral AI has deprecated `mistral-moderation-2411`, urging migration to `mistral-moderation-2603` for enhanced content policy.

Principles

• Regularly update moderation models.
• Refine policy categories for clarity.

Method

Configure custom guardrails using `moderation_llm_v1` by setting category thresholds and defining an action like "block" for content violations.

In practice

• Migrate `moderation_llm_v1` configs to `moderation_llm_v2`.
• Review new "Dangerous", "Criminal", "Jailbreaking" categories.

Topics

• Mistral Moderation 2411
• Mistral Moderation 2603
• Content Moderation
• Guardrail Configuration
• Policy Categories

Best for: CTO, VP of Engineering/Data, Director of AI/ML, AI Engineer, MLOps Engineer, AI Security Engineer

Related on AIssential

• Mistral's Le Chat spreads Iran war disinformation in 60 percent of leading prompts — AI chatbots like Le Chat are highly susceptible to disinformation, especially with leading or malicious prompts.
• CHILLGuard: Towards Fine-Grained Chinese LLM Safety Guardrail with Scalable Data Construction and Model-aware Preference Alignment — CHILLGuard offers a fine-grained Chinese LLM safety guardrail via scalable data construction and model-aware preference alignment.
• Meta’s deepfake moderation isn’t good enough, says Oversight Board — Meta's deepfake moderation is insufficient, requiring an overhaul of AI content labeling and detection.
• What the Eyes See, the LLMs Miss: Exploiting Human Perception for Adversarial Text Attacks — LLM content moderation systems are vulnerable to visually-based adversarial attacks that exploit human perception.
• Breaking: Trump asks the impossible of Anthropic — Preventing LLM jailbreaks is impossible due to their fundamental next-token prediction architecture.
• New Ministral 3 14B vs Mistral Small 3.2 24B Review — Larger parameter models often perform better on complex document parsing, even with higher quantization.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by mistral.ai via Google News.