Safety
Preventing AI systems from producing harmful, dangerous, or unsafe output.
Safety in AI refers to designing and operating systems to reduce the risk of harm, such as generating dangerous instructions, enabling violence, or producing content that puts users at risk. On the AIF-C01 exam, safety is examined primarily through Amazon Bedrock Guardrails, a Bedrock feature letting builders define denied topics, blocked words, and content filters that intercept unsafe outputs before they reach users. The key distinction is between safety and fairness: safety targets harmful outputs (weapons, self-harm, illegal acts), while fairness targets discriminatory outputs. Guardrails can help with both, but they address different responsible-AI dimensions.
PlayPrepHQ study notes are written and reviewed against primary exam sources. How we create & review content →