MoltCops Disclosure Policy

Who we are

MoltCops is an independent security research project focused on AI agent infrastructure — the supply chain, credential hygiene, and attack surface of autonomous agent systems. We operate a free static-analysis scanner (MoltShield) and conduct ongoing public-data research into leaked credentials in agent ecosystems.

Scope

Our research covers publicly accessible data: public GitHub repositories, gists, published packages (npm, PyPI), Hugging Face Spaces, and search-engine-indexed pages. We do not access systems behind authentication, scrape private platforms, or use discovered credentials to gain access to anything.

What we do when we find an exposed credential

1. We classify it offline. Wallet keys are assessed by deriving the address locally (pure elliptic-curve math) and reading public chain data. API keys are classified by format and context. No credential is ever transmitted, used for authentication, or tested against its provider.
2. We notify the owner. We use the project's official security channel first (SECURITY.md, private vulnerability reporting, security@ email). If none exists, we use the repository owner's public contact email. Notifications are sent from moltcops@proton.me and signed with our PGP key.
3. We differentiate by finding type. For exposed credentials (API keys, private keys, tokens, seed phrases), the fix is revocation and rotation — this takes minutes to hours, not months. We notify immediately and expect remediation within 72 hours. For software vulnerabilities (contract bugs, logic flaws, access-control issues), we follow the standard 90-day coordinated disclosure timeline to allow for code changes, testing, and deployment. We are flexible on this window for complex fixes. In both cases, we send one follow-up after 7 days if there is no initial response.
4. For crypto projects with funds at immediate risk and no responsive contact, we escalate to SEAL 911 (Security Alliance) for emergency routing.
5. We publish only aggregate, anonymized statistics. Never the credential, never the full repository name, never a link to the leak. Monthly reports contain counts by category, response rates, and time-to-remediation — nothing that identifies a specific project or individual. Because we publish only aggregate statistics and never identify specific projects or individuals, the publication timeline is decoupled from the remediation timeline.

What we never do

• Use a discovered credential for any purpose, including verification against its provider.
• Move, transfer, or interact with funds in any wallet, including "rescue" transfers.
• Access private systems, repositories, or data using a discovered credential.
• Demand payment, threaten publication, or condition disclosure on compensation.
• Publish findings about projects where we hold any financial position.

Bug bounty programs

Where a project operates a formal bug bounty (Immunefi, HackerOne, or a self-hosted program), we work within that program's scope and terms. We do not request compensation outside established programs.

Contact

moltcops@proton.me · PGP key at moltcops.com/pgp-key.txt · Fingerprint: D5E0 A36D 2F1F 2CCE 7055 B7F7 11A0 9D1B D499 CD37.

For projects that want proactive scanning

MoltShield (our free scanner) is available as a GitHub Action that checks pull requests for agent-security patterns. See github.com/moltcops-security.