Research program
Before society can responsibly make decisions about increasingly capable AI systems, it needs stronger foundations for understanding what those systems can do, how they should be evaluated, and how trustworthy evidence should be produced.
This page introduces the conceptual foundations that underpin the entire Standards Body project. It is not a technical standard, not a regulatory proposal, and not an implementation guide. It is a framework describing the foundational infrastructure that should exist before society attempts to make increasingly consequential decisions about frontier AI. These foundations are intended to evolve as evidence improves.
Public editions of all eight foundation papers, the overview white paper, and the project's foundational sources are available in the Library, each with its version, status, downloads, and correction record.
History repeatedly shows that durable institutions are rarely built by beginning with enforcement. They begin by creating shared language, shared measurements, shared evidence, shared processes, and shared expectations. Medicine required clinical methods before licensing systems. Engineering required measurement before building codes. Aviation required investigation before regulation. Cybersecurity required vulnerability disclosure before coordinated defense.
Likewise, frontier AI requires evaluation infrastructure before reliable governance. Without foundations, measurements become inconsistent, claims become difficult to compare, institutions lose legitimacy, incentives drift toward marketing rather than evidence, and decision-makers lack reliable information.
Standards Body therefore prioritizes foundational infrastructure before institutional authority.
Governance answers: what should be done? Evaluation answers: what is actually true? Governance without reliable evaluation risks making decisions based on incomplete or misleading information. Evaluation without governance leaves important decisions unsupported. The relationship is complementary rather than competitive. The project's emphasis on evaluation reflects the belief that trustworthy evidence is one of the strongest long-term public goods that can be created for frontier AI.
Foundation 1
Static benchmarks inevitably lose relevance as models improve. Evaluation systems should evolve alongside capability growth through continuously updated methodologies, adaptive testing, and ongoing validation.
Public essay derived from this foundation →Foundation 2
High-consequence evaluations require mechanisms that reduce benchmark leakage and gaming while preserving scientific validity. Secure, independently managed evaluation resources become increasingly important as capabilities grow.
Read the foundation paper →Foundation 3
Some capabilities warrant greater scrutiny because of their potential societal impact. Evaluation infrastructure should distinguish between ordinary capabilities and domains where evidence quality must be substantially stronger.
Read the foundation paper →Foundation 4
Trust increases when evaluation incorporates diverse, technically qualified, independent expertise. Independent review strengthens credibility while reducing institutional blind spots and conflicts of interest.
Read the foundation paper →Foundation 5
Long-term evaluation capacity is unlikely to scale through developers alone. An ecosystem of qualified, accountable, independent evaluators can improve resilience, specialization, and public confidence.
Read the foundation paper →Foundation 6
Institutional expectations rarely emerge all at once. Voluntary practices may gradually mature into industry norms, procurement expectations, insurance requirements, certification mechanisms, or other structured forms over time.
Read the foundation paper →Foundation 7
Organizations respond to incentives. Recognition, credibility, reputation, transparency, and demonstrated excellence should increasingly reward participation in rigorous evaluation rather than mere claims.
Read the foundation paper →Foundation 8
Frontier AI development is international. Evaluation infrastructure should strive for compatibility across jurisdictions while respecting regional differences and encouraging collaboration instead of fragmentation.
Read the foundation paper →None of the eight foundations stands alone. Dynamic evaluations require independent expertise. Independent experts require accreditation. Accreditation benefits from transparent standards. Standards require trustworthy evidence. Evidence depends on robust evaluation. International interoperability depends upon shared terminology, shared processes, and comparable evidence. The framework should therefore be viewed as an interconnected system rather than eight isolated ideas.
No. One of the project's core principles is that foundational ideas themselves should remain open to revision. Future research may justify splitting a foundation into multiple domains, merging overlapping concepts, introducing additional foundations, retiring obsolete approaches, or reordering priorities. Foundations should earn permanence through evidence rather than assumption.
The current foundation papers are foundational research and working institutional proposals, not binding standards.
The first public derivative of this program is now available:
Static Benchmarks Are Not Enough: Why Frontier AI Evaluation Must Be Continuously Maintained →Public Foundational Essay · SB-PUB-2026-0001 · Version 1.0 · Published July 17, 2026
