Please report security issues privately — do not open a public issue for a vulnerability.

• Preferred: use GitHub's private vulnerability reporting on this repository (the Security tab → Report a vulnerability), which opens a draft advisory visible only to maintainers.
• We aim to acknowledge a report within a few days and to provide a remediation plan or fix timeline once triaged.

When reporting, please include the affected version, reproduction steps, and the impact you observed.

zing is a local-first auditing tool. The security properties that matter most:

• Credential handling. API keys are fingerprinted (SHA-256, truncated) and never stored verbatim. Reports and logs route every relay-controlled string through the redactor (zing/utils/redact.py) before serialization. A path that lets a configured key or another secret reach a JSON/Markdown/HTML report is an in-scope vulnerability.
• Untrusted input. Relay responses are untrusted by design. Report renderers must neutralize relay-controlled text (HTML-escape, Markdown-escape) so it cannot inject markup or spoof report structure.
• No surprising egress. zing only contacts the endpoints you configure (the target, an optional baseline, and an optional judge). A change that sends data elsewhere is in scope.

zing reports black-box evidence of divergence and risk, not proof of fraud. It cannot prove a provider logs prompts, always routes to one exact model, or commits billing fraud. Publishing an accusation against a vendor based on a single run is a misuse of the tool, not a security issue — see the "Responsible use" section of the README and docs/METHODOLOGY.md.