IGL is a governance and audit runtime. A security bug in IGL can break the identity, boundary, intent or trace guarantees that other systems rely on, so we treat every report seriously and quickly.

Do not open a public GitHub issue for security vulnerabilities.

Report privately through either channel:

• GitHub private vulnerability reporting: Report a vulnerability (preferred, creates a private advisory thread)
• Email: security@id-me.ai

Include a description of the vulnerability and which guarantee it affects (identity, boundary enforcement, intent recording, trace integrity, provenance rules), steps to reproduce or a proof of concept, and affected versions if known.

• Acknowledgment within 48 hours.
• Severity assessment within 5 business days, with a remediation timeline.
• We coordinate disclosure with you. Our target is a fix and advisory within 90 days; we will tell you if an issue is more complex and why.
• After a fix ships, we credit reporters in the advisory (unless you prefer anonymity).
• Advisories are published as GitHub Security Advisories with a CVE where applicable.

Issues that break a core guarantee are critical:

1. Executing work without a valid identity binding
2. Executing an action outside a declared boundary
3. Forging, altering, or truncating a trace without breaking hash-chain verification
4. Reclassifying provenance (making a lower class appear as granted or observed: governing)
5. Bypassing canonicalization so that two different payloads hash identically

Issues in the platform contract (spec/) that would allow a conforming client to bypass a gate are treated as high. Issues in examples or docs are welcome but lower priority. The hosted platform at id-me.ai is in scope for reports; its infrastructure is out of scope for testing without written authorization.

The October 2026 review (docs/audit/SECURITY-AUDIT-2026-10.md) lists nine findings. F-01 to F-04 were fixed in 1.3.0 and F-05 to F-09 in 1.5.0; each has a regression test (test/security.test.js, test/audit.test.js) and the probe scripts that reproduced them are kept in docs/audit/ and now show the properties holding. The review was conducted by the project itself. No independent review has been performed yet, and we will say so until one has.

• Release tags are signed. Verify a tag before building from source: git tag -v v1.5.0.
• Published packages carry npm provenance attestations; verify with npm audit signatures.
• You can check a trace yourself without trusting this project. A journal (MemoryJournal, FileJournal or D1Journal in src/store.js) has verify() for the hash chain, verify({ checkpoint }) to detect a truncated or rewritten history against an anchor you hold, merkleRoot() and inclusionProof(seq) for selective disclosure, and the static MemoryJournal.verifyInclusion({ leaf, proof, root }) for the auditor's side. Receipts, grants and approvals verify through Signer in src/sign.js. Everything is standard SHA-256 and Ed25519, so you can recompute it with any library you prefer.