Know what you received.
A signed manifest binds the model files and release documents. A SAI Bill of Materials (SAIBOM) records declared upstream components, licenses, data provenance and pipeline references.
Secure Super AI General Public License
A model can change hands. Its provenance, disclosed risks and security obligations should follow. SAIL-GPL makes release evidence part of the license.
For SAI (Super AI) model artifacts and their derivatives.
A license proposal.
A working evidence toolkit.
A signature can show that files are intact. A responsible release also explains where the model came from, what was evaluated and how to report a vulnerability. SAIL-GPL proposes that this evidence follows every distribution.
A signed manifest binds the model files and release documents. A SAI Bill of Materials (SAIBOM) records declared upstream components, licenses, data provenance and pipeline references.
The model card and signed evaluation disclose intended use, limitations and test results. Failed tests can be reported honestly. A valid signature never turns a weak model into a safe one.
Fine-tuning, merging and quantization create a new release obligation. Refresh the evidence, record the parent signature, sign the derivative and preserve coordinated disclosure duties.
Secure-Release Metadata / SRM
The draft couples permission to distribute a model with a duty to ship its evidence. The toolkit handles the technical record. The steward remains responsible for truthful declarations and appropriate evaluation.
Collect the model files. Write a meaningful model card and publish a disclosure contact and process.
Supply the SAIBOM and actual adversarial evaluation results. Choose the tier that fits the model distribution and its risk.
Bind the files and evidence to a release manifest. Keep private keys outside the distributed artifact.
Check the release and pin trusted keys obtained separately. For a derivative, renew the record and repeat.
SAIL-GPL covers model artifacts and derivatives. Independently licensed software retains its own terms. This draft does not make a claim of compatibility with another software license.
Explore how the release requirements change from a narrow model release to a high-risk distribution. All three tiers require the core evidence bundle.
T0 covers internal research that is not conveyed to third parties. Distribution triggers at least T1. Applicable law may impose further obligations.
For small or narrow models distributed to third parties. The evaluator may self-report, with explicit results for the minimum test set.
Without a trusted publisher key, verification establishes integrity only. Any supplied receipt requires a trusted log key.
The open Python toolkit builds and verifies native SAIL SRM v2 bundles. Clone the repository, install the toolkit dependencies, then try the signed demonstration release.
PYTHONPATH=sail/src python -m sail verify example-model --trusted-key sail/examples/example-public-key.json
The example key is published for a reproducible demonstration. In a real release, obtain and authenticate the publisher key through a trusted channel outside the model bundle. Default T1 and T2 verification checks integrity without establishing publisher identity. T3 also requires distinct trusted evaluator and log keys. Endpoint availability, truthful claims and operational disclosure handling remain outside the verifier's technical scope.
A public draft, open for scrutiny
Review the license language, examine the wire format or test the toolkit. SAIL-GPL 1.0 is a proposal for community and counsel review. It has no OSI approval or registered SPDX license identifier.