Please report security issues privately via a GitHub security advisory on sekacorn/ModelSwapBench rather than a public issue. We aim to acknowledge within a few days.

ModelSwapBench is local-first and privacy-preserving by default:

• No hosted providers by default. Benchmark data never leaves the machine unless you pass --allow-hosted and set privacy.allow_hosted_providers.
• No telemetry, no auto-upload, no account.
• Safe YAML. Benchmark files are parsed with yaml.safe_load; arbitrary Python is never executed. No custom-code evaluators in v0.1.
• Secret hygiene. API keys are read from environment variables by name and are never written to results or reports. Logs and reports are redacted.
• Path safety. All run artifacts are written within the storage directory; traversal and symlink escapes are blocked. Deletion is scoped and confirmed.
• Bounded resources. Concurrency, output size, retries, and timeouts are capped to mitigate runaway cost and denial-of-service.
• Untrusted output. Model output is treated as untrusted and HTML-escaped in reports.

See THREAT_MODEL.md for the full analysis.

v0.1.x (beta) receives security fixes on a best-effort basis.