Morphis is an agent OS that runs on user machines and executes code on the user's behalf. Security is a first-class concern, not an afterthought. We take vulnerability reports seriously.

If you believe you've found a security issue, please follow responsible disclosure below — do not open a public issue.

Report security issues privately to:

• Email: security@morphis.org (placeholder until The Nerfed Lab's security inbox is set up; in the interim, contact a current maintainer privately via a verified channel)
• PGP key: .well-known/pgp-key.asc — fingerprint coming when the Foundation is formed.
• Web form: https://morphis.org/security (coming soon)

Please include:

1. A clear description of the issue and its impact.
2. Reproduction steps — exact commands, configs, payloads.
3. Affected versions — Morphis version, OS, host LLM provider, relevant skill versions.
4. Suggested mitigation if you have one.
5. Your contact info so we can follow up and credit you.

We aim to acknowledge within 48 hours and provide an initial assessment within 7 days.

• Morphis core (Agent Kernel, Permission Broker, Sandbox, LLM Router, Critic Agent, Self-Modification system)
• Living Vault memory subsystem
• Voice pipeline (wake / STT / TTS)
• Official skills shipped under the @morphis/ namespace
• The morphis.policy schema and signed-update pipeline
• The Morphis daemon and CLI

• Third-party skills not shipped under @morphis/ — report directly to the skill maintainer; we will help coordinate.
• Vulnerabilities in upstream dependencies (Claude API, LiteLLM, Matterbridge, LanceDB, Obsidian, etc.) — report to the upstream project; we will track.
• Issues in user-owned forks not endorsed by The Nerfed Lab.

The Morphis threat model is documented in detail at docs/researchers/safety-model.md. Key assumptions:

• Prompt injection is always possible. Skills receive untrusted user content via voice, files, network, and screen. Every skill is sandboxed.
• Malicious skills will be submitted to the registry. The Lab tier requires review + signing + sleeper delay before high-capability skills can use trusted credentials.
• Self-modification proposals can be adversarial. The §12 critic agent runs in a separate process on (where feasible) a different LLM provider, and L5 changes require a two-person rule.
• The user trusts Morphis on their machine but not necessarily everything Morphis is asked to do. All destructive or irreversible actions require explicit approval.

• Day 0: report received.
• Day 0–2: acknowledgment.
• Day 0–7: initial assessment, severity classification.
• Day 0–90 (typical): fix developed, tested, prepared for release.
• Coordinated release: patch ships; advisory published; reporter credited (unless they prefer anonymity).
• +90 days from patch: full technical details may be published.

We follow Project Zero–style 90-day disclosure for most issues. Critical issues affecting user safety may be patched faster.

Reporters are credited in the security advisory and in the release notes (with permission). We are working toward a bug bounty program funded by The Nerfed Lab; until then, recognition is the reward, and significant findings will earn The Nerfed Lab's acknowledgment.

None yet (pre-V0 project). Advisories will be published at https://morphis.org/security and in docs/researchers/advisories/ once the project ships.