As AI agents take over increasingly complex workflows, enterprises demand validation that they are secure, safe and reliable. Weekly headlines of AI failures - from chatbots hallucinating to data leaks - make the need for third-party validation of robustness even stronger. New modalities, such as voice AI, require adaptation of standards to meet unique risks.

AIUC-1 was developed to meet this demand for agentic AI specifically and ElevenLabs is now the first AI voice company to achieve AIUC-1 certification, meeting the standard’s 51 requirements spanning technical controls, legal policies, and adversarial testing.

AIUC-1 is developed with Technical Contributors from trusted organizations including Cisco, CSA, Orrick, and MITRE. The standard addresses the six risk categories enterprises evaluate before AI adoption: data & privacy, security, safety, reliability, accountability, and society.

Testing voice-specific risks

Unlike traditional AI governance frameworks, AIUC-1 requires regular technical testing of AI agents to validate their robustness to risks like hallucinations, harmful outputs and prompt injections. To obtain certification, ElevenLabs agents were subjected to 5,835 technical tests covering 14 risk categories. 

The testing included voice-specific risk surfaces such as maintaining reliability across languages (with agents subjected to prompts in both English, Spanish and Hindi), system robustness to accents and muffled voices, and transformation-attacks including background noise injection and pitch modulation.

ElevenLabs customers can streamline certification of their own voice agents

Beyond announcing the AIUC-1 certification of ElevenLabs’ own agent, ElevenLabs can also announce that customers building and deploying agents on the ElevenAgents Platform are a step closer to AIUC-1 certification. 

By integrating a comprehensive suite of safeguards and guardrails directly into the platform, agents built on the ElevenAgents Platform are up to 75% of the way towards certification. For example, several AIUC-1 safety requirements such as C003: Prevent harmful outputs are met by built-in safety guardrails that equip agents with defensive prompting and harmful content filtering. Similarly, a built-in jailbreak classifier meets AIUC-1’s B002: Detect adversarial input.

The result is that ElevenLabs customers typically can achieve the full AIUC-1 certification in weeks.

Supporting the development of voice-specific requirements

In addition to obtaining AIUC-1 certification, ElevenLabs joins AIUC-1 as a Technical Contributor to share insights on voice-specific risks and emerging safeguards. ElevenLabs joins as a Technical Contributor alongside contributors from trusted institutions including Stanford’s Trustworthy AI Research Lab, MITRE, Cisco, MIT, and the Cloud Security Alliance, who work collectively to shape AIUC-1 requirements for new agent modalities like voice AI. 

New, voice-specific controls have already been added to requirements including E016: Implement AI disclosure mechanisms and C003: Prevent harmful outputs in the January 15th, 2026 standard update, with additional voice-specific controls being released at the next update on April 15th, 2026.