A curated, license-tagged reference catalog of open standardized legal agreement frameworks. Every mature open legal standard converges on the same two-layer shape: a fixed, review-once body incorporated by reference, plus a cover page of typed variables. This directory maps the landscape of those frameworks, covering both computable open standards ready for assembly (Strand A) and reference-only standards (Strand B).
The 2024-2026 wave of AI deployment has produced a first cohort of open agreement standards specifically addressing AI data use, procurement, and regulatory compliance. These are the most directly relevant to the AINumbers suite's AI-Act and agent-commerce tooling.
Three frameworks land squarely where this suite is strongest: binding AI-Act risk classification to procurement clauses and agent-readable terms.
Cover-page variables: AI data-use permissions, training opt-out, model improvement, retention window, output ownership, subprocessor AI controls. Pairs with Common Paper agreements. Suite cross-links: TDM reservation (art-202), AI Act Art 50 marking (art-126), High-risk fit classifier (art-64). First-class assembler target for the P2.5 AI-vendor onboarding workflow.
European Commission-supported procurement clause sets for AI Act Chapter III high-risk AI systems. Covers transparency, risk management, data governance, human oversight, and cybersecurity. Redistribution terms unclear: map and point, never vendor. Suite cross-links: AI-Act high-risk classifier (art-64), agentic AI risk classifier (art-67), Art 50 marking check (art-126). Target for the P2.5 EU procurement clause mapper.
Mandatory clause for US federal AI system procurements under the General Services Administration acquisition regulations. Covers AI transparency, risk management, and incident reporting obligations. Reference-only: available at acquisition.gov. Suite cross-link: AI-Act high-risk classifier (art-64).
Strand A: computable frameworks whose cover-page variables can be bound to a kernel and whose body is deterministic. Strand B: human-prose standards that are reference-only, not nodes. The immutable-body model applies to all: bodies are never modified regardless of license. All variability lives in the cover-page layer.