Landscape

Where AIIR fits among AI provenance and supply-chain attestation formats.

Context

Several formats address software supply-chain integrity and AI provenance. They serve different purposes and often complement each other. This page presents factual capabilities — not opinions — so you can evaluate which formats apply to your use case.

Where possible, links point to the primary specification or repository for each format. All information was last verified March 2026.

Feature comparison

Capability AIIR in-toto SLSA SPDX CycloneDX Sigstore
AI authorship detection
Content-addressed receipts
Zero runtime dependencies
Binary wire format (CBOR)
Normative specification
Published test vectors✓ (49)
Cross-language conformance✓ (Python + Rust + JS)
Public threat model✓ (STRIDE/DREAD, 147+ controls)
Continuous fuzzing✓ (Hypothesis + ClusterFuzzLite)
Mutation testing
Browser-based verifier
Sigstore integration
SBOM / dependency graph
Build provenance attestation
EU AI Act evidence

✓ = capability present in the format or reference implementation. — = not a design goal or not present. Comparisons based on public documentation as of March 2026.

Complementary, not competing

These formats address different layers of the supply-chain trust stack. They are designed to work together:

LayerQuestion it answersFormats
AI authorship Was AI involved in this commit? How much? What kind? AIIR
Build provenance Where was this artifact built? By whom? From what source? in-toto, SLSA
Dependency inventory What components does this software contain? SPDX, CycloneDX
Signing & transparency Can we verify the signer's identity without managing keys? Sigstore

AIIR receipts can be wrapped as in-toto predicates (predicate type: https://aiir.io/commit-receipt/v1), and AIIR signs with Sigstore natively. An enterprise deployment might use AIIR for AI provenance, SLSA for build provenance, SPDX for SBOMs, and Sigstore for signing — each format in its own lane.

Why this matters

The EU AI Act (Article 52) and SOC 2 Type II both require documented evidence of AI involvement in production systems. No existing supply-chain format provides this evidence natively. AIIR was designed specifically for this gap: a content-addressed, tamper-evident receipt that records what AI did, when, and how much — at the commit level.

The format's zero-dependency design means it can be adopted without introducing new supply-chain risk. The published test vectors and multi-language reference implementations mean any engineering team can build a conforming verifier without relying on a single vendor.

Evaluate for yourself

The spec, test vectors, and threat model are all public. Audit them.

Read the spec → Conformance guide →