Guardrails outside the model: authorizing AI agent actions
Why guardrails inside an LLM can be talked around, and how a deterministic decision made outside the model limits what prompt injection can change.
Read the full post: https://sebastianrenker.github.io/blog/guardrails-outside-the-model/
Originally published at https://sebastianrenker.github.io/blog/guardrails-outside-the-model/.
原文：https://www.a8q9.cn/en/article/guardrails-outside-the-model-authorizing-ai-agent-actions-4ca531.html