ElevenLabs rolls out Guardrails 2.0

ElevenLabs launches Guardrails 2.0 in alpha for ElevenAgents, adding three levels of real-time protection, Custom Guardrails, and Zero Retention data mode.

ElevenLabs is launching Guardrails 2.0 in alpha within ElevenAgents, its AI voice agent infrastructure. The system relies on three levels of real-time protection: system prompt reinforcement (Focus), detection of user-side prompt manipulation or injection attempts, and validation of agent responses before output. Complementing these pre-integrated components are Custom Guardrails, which allow defining business rules in natural language, automatically evaluated with each interaction. Together, these offer granular control over execution modes (latency vs. rigor), actions triggered in case of violation (conversation termination, transfer, human escalation), and sensitivity thresholds per content category. ElevenLabs also integrates automatic redaction of sensitive data in post-call transcripts and recordings, with a Zero Retention mode for deployments subject to strict compliance requirements.