If you discover a way to bypass constitutional invariants, defeat epistemic marker enforcement, produce false receipts, or otherwise compromise the adapter's governance guarantees — we want to know immediately.

Do not open a public issue.

Contact Stephen Hope directly via LinkedIn: https://www.linkedin.com/in/stephen-hope-75497937a

Include:

• A description of the attack vector
• Steps to reproduce
• Affected component (prompt, markers, marker coverage, receipt, API)
• Suggested fix (if you have one)

We will respond within 48 hours and coordinate a fix. You will be credited publicly unless you request otherwise.

This policy covers:

• Constitutional bypass — any method that causes the adapter to drop or circumvent epistemic markers
• Marker-coverage evasion — any method that causes the adapter to report an incorrect drift_score
• Receipt tampering — any method that breaks or falsifies the cryptographic receipt chain
• Authority spoofing — any method that convinces the adapter to accept inline constitutional amendments
• API exploitation — any vulnerability in the FastAPI reference server or Foundry
• Tenant isolation / IDOR — any method that allows one node to access another node's sessions, ledger entries, or receipts
• Key management — bypass of the X-API-Key auth gate, extraction of key material, or forgery of stored hashes
• Rate limit bypass — resource exhaustion via forwarded-header spoofing or unbounded state accumulation
• Widget API abuse — unauthenticated access to model-backed endpoints or CORS policy bypass

We support coordinated disclosure. Give us reasonable time to patch before going public. We'll treat your report seriously and move fast.

Past contributors to adapter security:

• anitgravity — adversarial audit (v1.2), codebase architecture review
• Pliny the Prompter — jailbreak toolkit used in red-team testing (GODMODE, Parseltongue)
• Fable 5 — API security audit (v1.7.0): key hashing at rest, tenant isolation / IDOR, rate limiter trust-proxy fix, widget CORS and auth gate