Miles Brundage, former policy lead at OpenAI, has launched AVERI, a new nonprofit organization focused on advancing independent third-party auditing of frontier AI systems.

Brundage left OpenAI in October 2024 during the company’s restructuring of its AGI Readiness team, arguing at the time that no frontier AI labs were adequately prepared on the safety front for advanced systems.

Rather than acting as an auditor itself, AVERI positions itself as a policy and standards organization. Its goal is to define audit frameworks, assurance levels, and governance models that could later be used by independent evaluators, regulators, or international bodies.

The organization has received early donations from employees across leading AI labs. Brundage has said these contributors “know where the bodies are buried,” underscoring insider concern about the limits of current self-reported safety testing.

AVERi has also published a framework outlining four “AI Assurance Levels,” ranging from today’s self-testing regimes (Level 1) to treaty-grade international verification mechanisms (Level 4).

Why it matters: The AI industry has largely relied on voluntary disclosures and internal evaluations to assess safety. AVERI’s insider-backed push for external scrutiny could mark an early shift toward formalized third-party accountability as frontier models grow more powerful.