Security & abuse model¶
In scope:
- Replay poisoning: mitigated via provenance hashes and immutable artifacts.
- Artifact spoofing: mitigated via fingerprint checks and audit chain.
- Capability refusal: ND without ANN results in
NDExecutionUnavailableError. - Authz denial: mutations blocked via
Authzcontracts.
Out of scope:
- Multi-tenant isolation guarantees.
- Network-level ACLs beyond Authz hooks.
- Side-channel attacks or timing analysis.
Operator guidance:
- Treat provenance ledger as tamper-evident; rotate or archive per retention policy.
- Run divergence detection to catch backend drift.
- Do not bypass invariants via retry wrappers.
- Vulnerability posture:
py==1.11.0triggers PYSEC-2022-42969 with no upstream fix yet. SBOM generation ignores this ID explicitly until a patched release is available; upgrade once released.