UST is trust infrastructure: the reference verifier/checker (checkAuthorityProofBytes and the resolveKeys / resolveAuthority / verify / verifyAsync / resolveByDiscovery name-binding path) is the TCB — a soundness or totality defect there is treated as critical. The specification and its measure-theoretic formal model (spec/UST-1.0-formal-model.md) are part of the security surface: a claim the model asserts but the code does not realize is a security issue, not a documentation issue.

Report privately — do not open a public issue for a suspected vulnerability.

• Preferred: GitHub private vulnerability reporting — the repository's Security → Report a vulnerability tab (opens a private advisory only maintainers can see).
• A minimal, runnable reproduction against the exact commit is the most useful report: bytes/inputs in, the verdict out, and the invariant you believe is violated (cite the spec/ or formal-model passage where possible).

We aim to acknowledge a report within a few days. Because the project is small, please allow reasonable time to remediate structurally before any public disclosure; we will credit reporters who wish to be named.

• The reference verifier and the name-binding resolver (the TCB above) returning VALID(J) / authoritative / anchored for an input with no genuine derivation of J; any way to make it throw, loop, or return a non-deterministic / non-tri-state result; or any cross-implementation divergence at the canonical byte boundary.
• The conformance vectors (vectors/) or the spec/model being unsound or internally inconsistent.

• The untrusted object-encoding adapter checkAuthorityProof(obj) (only the immutable-byte …Bytes boundary is the TCB), and the producer/prover stack.
• Operational deployment of a specific operator (that is the operator's own security surface).

Active development is on the 1.0.0-rc.x line; fixes land on main. There is no long-term-support branch yet — pin an exact version and track main for the current state.