Safety Audits

Independent certification for embodied AI systems

Launching 2027

Safety certification services are currently in development. The Multi-Agent Safety Standards framework is being validated with industry partners. We expect to offer commercial certifications in Q2 2027.

Register for early access

What We Will Certify

Third-party safety assessment for humanoid robots and Vision-Language-Action (VLA) systems operating in human environments. Certification validates adversarial robustness, multi-agent safety, and failure recovery capabilities against evidence-based standards.

Certification Framework

Adversarial Robustness

  • Grounded in a 141,691-prompt jailbreak corpus across 231+ models
  • VLA-specific attack scenarios (visual adversarial patches, action-space perturbation)
  • Multi-turn interaction resilience testing
  • Quantified success rate thresholds by severity class

Multi-Agent Safety

  • Environment shaping resistance
  • Delegation cascade failure modes
  • Narrative erosion detection capabilities
  • Inter-agent trust calibration

Failure Recovery

  • Human intervention mechanisms
  • Graceful degradation paths
  • Reentry support after adversarial input
  • Logging and audit trail completeness

Regulatory Alignment

  • Australia VAISS Guardrail 4 compliance (pre-deployment testing)
  • EU AI Act Article 9 compliance evidence
  • NIST AI RMF function mapping
  • ISO/IEC 42001 control coverage
  • NSW WHS Digital Work Systems Act alignment
  • Insurer risk assessment compatibility

Certification Levels

Three-tier system (Bronze/Silver/Gold) based on adversarial success rate thresholds, recovery capability maturity, and audit evidence completeness. Certification is valid for 12 months and requires annual re-assessment.

Target Market

Development Partners

We are seeking 3-5 design partners to validate the certification framework. Partners receive free pilot certification and ongoing consultation in exchange for feedback on standards development.

Apply as Design Partner

Updates

Framework development updates are published in the policy brief series. Subscribe to the blog for monthly progress reports.