What the standards body will do
The new organization will establish requirements for independent third-party testing of AI models before deployment, define incident reporting standards for AI safety and security events, create auditor qualification criteria for organizations conducting model assessments, and codify voluntary safety commitments made by the participating labs. The effort builds directly on the 2023 Frontier Model Forum, a consortium that produced initial voluntary commitments but lacked a formal enforcement or verification mechanism.
Why this matters for security practitioners
Independent model auditing infrastructure is the prerequisite for any enforceable AI security baseline. Without a qualified third-party auditing ecosystem, claims about model safety and security properties cannot be independently verified. The incident reporting standard is particularly significant: it would create a mechanism for aggregating AI safety and security failures across deployments, analogous to what CISA's Ransomware Vulnerability Warning Pilot does for infrastructure vulnerabilities. For practitioners deploying AI in production, this body's output will eventually become the reference point for vendor due diligence questionnaires and procurement standards.
Context: What came before
The Frontier Model Forum (2023) brought the same core labs together but produced only voluntary commitments with no independent verification structure. A subsequent public-private partnership attempt within the Trump administration stalled, leaving a regulatory vacuum that industry is now filling itself. The EU AI Act created mandatory requirements for high-risk AI systems in Europe but lacks the granular technical standards for model-level security testing that practitioners need. This new body is positioned to produce those technical standards.
Concerns and limitations
Critics note the body could effectively exclude open-source model developers and smaller AI companies from its governance structure, creating a two-tier system where frontier model labs self-regulate while the broader AI ecosystem operates without equivalent scrutiny. Progressive Robot and other observers have flagged potential antitrust concerns given the degree of coordination among direct competitors. The body's independence from its founding members will determine whether its standards have credibility: a standards body funded and staffed by the same companies it audits faces inherent conflicts of interest.
What to watch
Watch for formal launch announcements expected by end of 2026 or early 2027, initial published standards and testing frameworks, and whether open-source model developers and non-signatory AI companies are granted observer or participation rights. The first auditor qualification criteria and incident reporting templates will be the most practically useful outputs for security teams.
Gigia Tsiklauri is a Security Architect and founder of Infosec.ge. Get in touch if you want to discuss this post or our security research.