Guardrails
Controls that constrain what an AI system accepts and produces.
Guardrails filter inputs and outputs (for harmful content, personal data or off-topic requests), enforce allowed actions, check outputs against sources and require approval for sensitive actions.
On the exam
Guardrails reduce but don’t eliminate risk; they must be tested and monitored like any other control.
Related terms
- JailbreakingTricking a generative AI system into ignoring its safety rules.
- Prompt injectionText that tricks a language model into ignoring its instructions.
- Human in the loopA person reviews or can override an AI system’s decisions.
AAIA Test yourself on this with timed AAIA practice questions in the AAIA Exam Simulation Hub, or read the free AAIA exam guide.
Test yourself with today’s question of the day.
