Glossary / Guardrails

Guardrails

Rules and filters built around an AI model to block harmful, unsafe or off-topic outputs.

Also written as: guardrail

In the news

• GLM 5.3 built Firefox exploit chains in patch-gap testsOct 30
• Mistral AI releases weights for improved Mistral Large 4Oct 29
• Army Cyber Command is developing self-spreading AI agents for cyberattacksOct 20
• Microsoft details MAI-Base-2 with 2 trillion parametersOct 17