Skip to content
/ aixv Public

AIXV helps organizations answer high-stakes questions before deploying or accepting AI artifacts

License

Notifications You must be signed in to change notification settings

AIXV-org/aixv

AIXV repository banner

AIXV - AI Integrity eXchange & Verification

CI Conformance CodeQL Scorecard Workflow PyPI version

AIXV is an open standard for AI artifact attestation, provenance, rollback, compromise detection, and investigation.

In practical terms, AIXV helps organizations answer high-stakes questions before deploying or accepting AI artifacts:

  • What exactly is this artifact?
  • Who produced or approved it?
  • What evidence supports trusting it?
  • Is it currently affected by an advisory or policy violation?
  • If compromised, what is the safest rollback path?

AIXV is built for three audiences that need shared, verifiable answers:

  • Technical teams: deterministic verification and machine-readable admission decisions.
  • Enterprise and public-sector risk owners: auditable evidence, policy controls, and incident traceability.
  • Policy, governance, and assurance functions: explicit trust assumptions, conformance checks, and compatibility contracts.

AIXV composes Sigstore cryptographic primitives and adds AI-native semantics:

  • Artifact typing,
  • Lineage graphs,
  • ML-specific attestations,
  • Advisory/recall workflows,
  • Policy-driven verification.

Release Posture

Current maturity: Pre-Alpha

This repository is a functional preview of the AIXV standard, but not yet a final ratified standard release.

Stable enough for pilot evaluation

  • Core primitives and schemas:
    • SignedRecord (aixv.signed-record/v1)
    • VerifyPolicy (aixv.policy/v1)
    • AdmissionDecision
  • Fail-closed verification and policy semantics.
  • Deterministic JSON output mode (--json) with tested contract behavior.
  • CI quality gates (ruff, mypy, pytest, build).

Still evolving

  • Broader conformance vector coverage and certification workflow.
  • Formal governance and external audit signals.
  • Wider ecosystem integrations and migration tooling.

Adoption Signals

For security and procurement reviews, the strongest immediate signals are:

  • aixv conformance --json produces a machine-readable conformance report.
  • docs/THREAT_MODEL.md, SECURITY.md, and docs/COMPATIBILITY.md define trust, reporting, and compatibility expectations.
  • CI enforces lint, typing, tests, build, and dedicated conformance workflow checks.
  • Scorecard and CodeQL workflows provide continuous security posture visibility in GitHub Security.

Standards and Security Docs

  • docs/AIXV_STANDARD.md
  • docs/NORMATIVE_CORE.md
  • docs/QUALITY.md
  • docs/THREAT_MODEL.md
  • docs/COMPATIBILITY.md
  • docs/TERMINOLOGY.md
  • docs/REGISTRIES.md
  • docs/ASSURANCE_LEVELS.md
  • docs/CONFORMANCE.md
  • docs/GOVERNANCE.md
  • docs/REPO_CONTROLS.md
  • SECURITY.md
  • RELEASE.md

Installation

pip install aixv

Quickstart (Core Flow)

# 1) Sign an artifact
aixv sign model.safetensors --identity-token-env SIGSTORE_ID_TOKEN

# 2) Create and sign a policy record
aixv policy create --input policy.json --sign

# 3) Verify artifact with signed policy + trusted policy signer
aixv verify model.safetensors \
  --policy .aixv/policies/policy.json \
  --policy-trusted-subject security-policy@aixv.org \
  --assurance-level level-2 \
  --json

# 4) Run conformance checks
aixv conformance --json

# 5) Optional: enforce signed-and-trusted attestations in lineage/export flows
aixv provenance model.safetensors \
  --require-signed-attestations \
  --trusted-attestation-subject ci-attestations@aixv.org \
  --json

CLI Surface

# Version
aixv version

# Signing
aixv sign model.safetensors --identity-token-env SIGSTORE_ID_TOKEN

# Verification
aixv verify model.safetensors --identity alice@example.com --issuer https://accounts.google.com

# Attestation
aixv attest model.safetensors --predicate training --input training.json

# Provenance
aixv provenance model.safetensors --depth 3
aixv provenance model.safetensors --view explain --depth 3 --json

# Advisory
aixv advisory create --advisory-id ADV-2026-0001 --severity critical --input advisory.json --sign
aixv advisory verify .aixv/advisories/ADV-2026-0001.json --trusted-subject security@aixv.org
aixv advisory sync --feed advisory-feed.json --trusted-subject security@aixv.org --max-bundle-age-days 30

# Policy
aixv policy create --input policy.json --sign
aixv policy verify .aixv/policies/policy.json --trusted-subject security-policy@aixv.org
aixv policy template --assurance-level level-2 --json
aixv policy migrate --input policy.json --to-assurance-level level-3 --max-bundle-age-days 30

# Record
aixv record create --kind waiver --record-id WVR-2026-01 --input waiver.json --sign
aixv record verify .aixv/policies/policy.json --kind policy --trusted-subject security-policy@aixv.org

# Bundle
aixv bundle create --input bundle.json --sign
aixv bundle verify .aixv/records/bundle/bundle-main.json --trusted-subject release@aixv.org

# Conformance
aixv conformance --json

# Rollback
aixv rollback model-v2.safetensors --to sha256:... --identity-token-env SIGSTORE_ID_TOKEN

# Export
aixv export model.safetensors --format in-toto
aixv export model.safetensors --format slsa --json
aixv export model.safetensors --format ml-bom --json

Policy Example

{
  "policy_type": "aixv.policy/v1",
  "allow_subjects": ["alice@example.com"],
  "allow_issuers": ["https://accounts.google.com"],
  "advisory_allow_subjects": ["security@aixv.org"],
  "max_bundle_age_days": 30,
  "deny_advisory_severity_at_or_above": "high",
  "require_no_active_advisories": false,
  "require_signed_advisories": true
}

Development

git clone https://github.com/aixv-org/aixv.git
cd aixv
python3 -m venv .venv
. .venv/bin/activate
pip install -e '.[dev]'

Quality Gates

ruff check .
ruff format --check .
mypy src
pytest
python -m build

About

AIXV helps organizations answer high-stakes questions before deploying or accepting AI artifacts

Topics

Resources

License

Contributing

Security policy

Stars

Watchers

Forks

Packages

No packages published

Languages