Do not open a public issue for security vulnerabilities.

Instead, please report vulnerabilities privately by emailing security@avarok.net with:

1. Description — What the vulnerability is and its potential impact
2. Reproduction steps — Minimal steps to reproduce the issue
3. Environment — OS, CUDA version, GPU model, Rust version
4. Affected component — Which crate or kernel is affected

We will acknowledge receipt within 48 hours and provide an initial assessment within 7 days.

Atlas is an inference server that runs locally with GPU access. The primary threat surface includes:

• CUDA kernel safety — Out-of-bounds memory access, buffer overflows in GPU kernels
• HTTP API — Input validation on the OpenAI-compatible endpoint (spark-server)
• Weight loading — Malicious safetensor files, path traversal during model loading
• Unsafe Rust — Atlas uses unsafe blocks for CUDA FFI; these are high-priority review targets

Atlas runs one automated security check in CI:

• cargo deny — Audits dependencies for known advisories, license compliance, and banned crates. Runs on every pull request, on pushes to main that touch Cargo.toml/Cargo.lock/deny.toml, and weekly. See .github/workflows/security.yml.

There is no automated static analysis of the CUDA kernel sources. Kernel memory safety is covered by human review and by the runtime kernel audit, not by a CI tool. This section previously advertised a cppcheck job that has never existed in this repository; claiming a control you do not run is worse than claiming none, because it tells a reader a class of defect is already being looked for.

We follow coordinated disclosure. Once a fix is available, we will:

1. Merge the fix to main
2. Tag a release
3. Credit the reporter (unless anonymity is requested)