Report suspected vulnerabilities privately through GitHub private vulnerability reporting or email security@thirdshift.example with Thirdshift security in the subject.

Include affected version, platform, reproduction steps, impact, and whether any data or credentials may have been exposed. Do not publish exploit details until a fix or mitigation is available.

Thirdshift v0.1 is not a trusted execution environment and is not confidential compute. Community hosts can control the machines that execute model inference. Use Thirdshift only for public or non-sensitive text workloads.

Do not send medical, financial, student, authentication, secret, regulated, enterprise-confidential, or other sensitive data through the alpha network.

Release zips are accompanied by SHA256SUMS and release-manifest.json. The manifest is signed with an Ed25519 release key, and thirdshift update verifies the manifest signature and artifact hash before promotion.

See docs/RELEASE.md for key generation, signature verification, and rotation notes.

• Host runtime binds to 127.0.0.1; nodes do not expose public inbound runtime ports.
• Node sessions are outbound WebSockets to the coordinator.
• Runtime and model artifacts are hash-checked before use.
• Job result envelopes are signed by node identity keys and verified by the coordinator before credit is created.
• Prompt and completion bodies must not appear in default logs.

The full threat model is in docs/THREAT_MODEL.md.

1. Triage and reproduce the report privately.
2. Disable affected manifests, release artifacts, API keys, or node accounts if containment is needed.
3. Patch and test the affected component.
4. Publish a signed release or documented mitigation.
5. Add regression tests and update the threat model or decision log if the boundary changed.