AVAAS Information

AVAAS™ certifies how your AI system behaves at the moment its output reaches a person and drives a decision that matters. Starting January 1, 2027, roughly 1,000 US organizations acquire deployer obligations under California and Colorado law. Illinois just became the first state to write independent third-party AI audits into law, starting with frontier developers in 2028 and requiring auditors with no financial conflicts of interest. The direction is set: regulators accept independent verification, not self-report. When your regulator, your insurance carrier, or your enterprise customer asks how you know your AI behaves, an internal eval is your own word. An AVAAS grade is independent proof. WHAT WE CERTIFY AVAAS-A evaluates the AI system itself. We test behavioral judgment under pressure, including whether an agent escalates to a human when it should, what it does after recognizing its own error, whether it resists manipulation, and whether it stays inside its mandate. AVAAS-D evaluates the deployment environment. We test whether the conditions around the agent, from escalation paths to permissioning, let good judgment actually operate. Every evaluation yields a grade you can show. A means Certified, B means Certified with Conditions, F means Not Certified, and Unassessable means we could see too little to grade. The grade is public-facing. Alongside it you receive a private remediation report with causal attribution, so your team knows exactly what to fix and why. WHY THE GRADE HOLDS UP The entity that holds the standard, the entity that runs the evaluation, and your organization are three separate parties. Re-tests are priced identically regardless of outcome. We cannot move the bar for you, which is precisely why the grade means something to the people you show it to. CONTINUOUS ASSURANCE Certification day and deployment day are different days. AVAAS Verified™ monitors live agents and measures whether production behavior matches evaluated behavior, pro