AI agent safety
Safety
A person can take the turn. A guardrail is a check you can fail. Recording notes on this site are educational and are not legal advice.
The packet
Diagram SVGDiagram PNGDiagram HTMLTrack map HTML
A human handoff in this lab means a person continues the turn. Pipecat's overview uses the word handoff for control moving to another agent. That is a different sentence.
Sort the packet
The handoff room names these four lines. Put them back in that order.
- Write the open question
- Write the note id
- Write what was said
- Write the channel
Rooms
- Handoff to a person, not a software handoff
- Guardrails
- Consent and recording, as education
- Evaluation and testing
- What this chapter will not claim
This chapter states no legal conclusion and configures no emergency calling.
Sources and go deeper
- Pipecat open source framework vendor last checked 2026-10
- NIST AI Risk Management Framework spec last checked 2026-10
- Anthropic, Building effective agents vendor last checked 2026-10
- OWASP Top 10 for Large Language Model Applications community last checked 2026-10
- Speech to Speech, docs.x.ai vendor The page title is Speech to Speech. The body names the Grok Speech to Speech API and also calls the site SpaceXSI Docs. last checked 2026-10
- 18 U.S.C. § 2511, Cornell Law School spec Educational reading of one federal section. Not legal advice, and not a survey of state law. last checked 2026-10
- LiveKit Agents, Pipeline types vendor last checked 2026-10
- Creative Commons Attribution-NonCommercial-ShareAlike 4.0 spec last checked 2026-10