Independent research and institutional design for frontier AI evaluation, assurance, standards, and governance.

Frontier AI is developing faster than the institutions that evaluate and govern it. Developers largely evaluate their own models, public benchmarks decay through contamination and saturation, and words like audit, certification, and accreditation are used interchangeably by people who need them to mean different things. Standards Body works on the missing infrastructure: shared terminology, evidence discipline, protocol governance, and credible third-party assessment, published openly so anyone can inspect, criticize, and improve it.

Standards Body is not a regulator, accreditation body, certification body, or governmental authority, and nothing it publishes is an adopted standard. That boundary is stated on every publication: standardsbody.ai/authority

Twenty-seven public records, every one versioned, statused, and correctable:

• Static Benchmarks Are Not Enough, the flagship essay: why consequential frontier AI evaluation must move from fixed benchmarks to versioned, governed, continuously maintained protocols. 18 minutes.
• A Crosswalk of Frontier AI Safety Frameworks: the evaluation triggers and commitments of the Anthropic, OpenAI, and Google DeepMind frameworks, side by side, maintained as they change.
• Eight foundation papers on dynamic evaluation protocols, held-out evaluations, high-stakes capability evaluation, independent review, third-party auditor ecosystems, progressive requirements, incentives, and global interoperability.
• Terminology explainers: what certification, accreditation, and audit actually mean, and why the differences matter for AI.

New here? Start with the short reading path.

Every released artifact ships with a published SHA-256, mirrored in this account's library repository and in the site's machine-readable catalog:

curl -s https://standardsbody.ai/library/catalog.json | python3 -m json.tool | head -40

Corrections are public, dated, and preserved: correction register. Released versions are never silently edited.

External review is open, and critical reviews are published identically to favorable ones: standardsbody.ai/engage. Error reports and contributions: standardsbody.ai/contact.

Standards Body is the work of Syed Hussain, independent and reader-supported. Funding principles and full disclosure: standardsbody.ai/support · @StandardsBodyAI