AXIOM is currently under active development.

Security fixes will generally target the latest version on the main branch.

Please do not open a public GitHub issue for a security vulnerability.

Instead, report security issues privately through the repository's GitHub security reporting mechanism.

When reporting a vulnerability, please include:

• a clear description of the issue
• affected component or feature
• steps to reproduce
• potential impact
• relevant logs or proof of concept
• any suggested mitigation

Please avoid including secrets, credentials, private datasets, API keys, or other sensitive information in the report.

Examples of security issues include:

• remote code execution
• arbitrary command execution
• authentication bypass
• privilege escalation
• unsafe model loading
• malicious model or dataset execution
• path traversal
• insecure API endpoints
• accidental exposure of credentials
• unauthorized access to private data
• vulnerabilities in AXIOM's deployment mechanisms

AI systems introduce risks beyond traditional software.

AXIOM may eventually process:

• downloaded models
• user-provided datasets
• generated files
• external repositories
• inference inputs
• training artifacts

Treat third-party models and datasets as untrusted inputs unless their origin and integrity are known.

Do not run untrusted model files or training code with unnecessary system privileges.

We aim to investigate valid security reports promptly and coordinate responsible disclosure where appropriate.

Once an issue has been fixed, relevant security information may be published through the repository's release notes or security advisories.

This policy applies to the AXIOM open-source project and its official repositories.

Third-party integrations, hosted infrastructure, and external services may have separate security policies.

Beta.5 fixes the reported model-path injection risks and administrator clear-text browser-storage issue. Report new vulnerabilities privately through GitHub security reporting. The public release record is in docs/releases/v0.2.0-beta.5.md.

Beta.5 hardens model-path handling against path traversal/path injection and removes administrator verifier material from clear-text browser storage. Report security problems through the repository security policy rather than public issue disclosure when sensitive exploitation details are involved.

Beta.5 includes hardening for untrusted model path handling and administrator browser storage.

Security issues should be reported privately through GitHub's repository security process.

The CodeQL path-expression findings reported in the model policy, optimizer, and model-path surfaces were reviewed on 2026-10-08. The confirmed runtime bundle issues were source-symlink disclosure and destination-symlink redirection in create_runtime_bundle. They were patched on 2026-10-08 by validating every resolved source and destination child, rejecting symlinks, and keeping runtime-profile writes below the trusted output root.

See the complete incident timeline, impact, patch, and regression coverage in SECURITY-ADVISORY-2026-10.md.