The DisposAI team takes the security of our software and its users very seriously. This document outlines our policy for reporting security vulnerabilities.

We provide security updates and patches for the following versions:

We recommend always running the latest release of DisposAI to ensure you have the latest security and stability fixes.

If you discover a security vulnerability in DisposAI, please do not open a public issue. Instead, follow one of these private reporting methods:

1. Go to the DisposAI Security Advisories tab on GitHub.
2. Click "Report a vulnerability" to submit a draft advisory privately to the maintainers.

If you cannot use GitHub Advisories, please reach out via GitHub private message or directly contact repository maintainers with the details.

To help us triage and resolve the issue quickly, please include:

• A clear description of the vulnerability and its potential impact.
• Step-by-step instructions or a minimal Proof of Concept (PoC) to reproduce the vulnerability.
• Affected component(s) (e.g. daemon-core, llama-backend, dispos-studio, SDKs).
• Environment details: Operating system, GPU architecture, CUDA version, DisposAI version.
• Any suggested fixes or remediations, if available.

1. Acknowledgment: We will acknowledge receipt of your vulnerability report within 48 hours.
2. Assessment & Confirmation: We will investigate and confirm the severity and affected components.
3. Patch Development: Maintainers will develop and test a security fix in a private branch or advisory fork.
4. Coordinated Disclosure: Once the patch is released in an official version update, we will publish a security advisory giving appropriate credit to the reporter (unless you request to remain anonymous).

DisposAI is designed to run local inference models on host hardware. Keep the following best practices in mind:

• Network Exposure: By default, daemon-core binds to 127.0.0.1 (localhost). Do not bind the daemon to public interfaces (0.0.0.0) on untrusted networks without proper authentication and TLS termination.
• Model Files: Only load GGUF or weight files from trusted sources (e.g. verified Hugging Face repositories). While GGUF is a safe tensor format, arbitrary pickle-based model formats should never be loaded from untrusted third parties.