Glossary
Guardrails
Checks placed around a model to block unwanted inputs or outputs, such as filters for sensitive data or policy violations.
Guardrails reduce risk but can be bypassed; they complement, rather than replace, careful permissioning of what the system can do.
Related terms
JailbreakPrompt injectionStaged release
Get the brief in your inbox
One email every morning: the stories that mattered in AI, each with a one-line "why it matters" for your role. Free, no ads, unsubscribe any time.
Published daily by AI Briefing. We never share your address.