Guardrails

Learn in five minutes

• OWASP's Top 10 for Large Language Model Applications is the community list this lab uses for model risks. It is not a contact-center certificate.
• NIST's AI Risk Management Framework is the vocabulary for measuring and managing that risk. It is not a product approval.
• A tool schema can make a request parseable. It does not make the request allowed.
• The xAI Speech to Speech page lists function as a tool type. A function that changes an account still needs a person when the action is hard to undo.
• Keep a step cap. The voice channel does not remove it.

Parseable is not permitted. Write the cap next to the tool name.

This section HTML

Visual

Diagram SVGDiagram PNGDiagram HTML

Flow SVGFlow PNGFlow HTML

One tool, one cap

1. Tool name: read_note.
2. Allowed argument: a note id from the call.
3. Cap: three steps, then the handoff packet.
4. Anything that charges or deletes is not in the list.

This section HTML

Quiz

This section HTML

Ecosystem

Guardrail

A check you can fail: schema, cap, and a person for an action that is hard to undo.

Sisters: Human handoff, NIST AI RMF, OWASP LLM risks, Recording statutes.

The voice channel does not remove the tool cap or the person who must approve an action that is hard to undo.

A later test can add a script you wrote. It cannot add a legal conclusion.

Config

• Write the goal in one sentence.
• Name the input and the allowed output.
• Save the trace next to the result.

Troubleshooting

• The answer looks fluent and is wrong. Check whether a source passage actually contains the claim.
• The loop never stops. Set a step cap and stop when the tool result repeats.

This section HTML