ĀML is a working research prototype with compiler, policy, receipt, interoperability, trust, and official-authorization surfaces. Security review is encouraged.

For a potentially exploitable security issue, please use a private channel first.

Contact: Office@aruintelligence.com

Please do not post active private keys, confidential customer information, exploit details that create immediate harm, or unreleased credential material in a public issue.

This repository does not currently promise a bug bounty or payment for reports unless a separate written program explicitly says otherwise.

High-value areas include:

• malformed-input crashes
• parser confusion
• output injection
• unsafe HTML emission
• compiler-state corruption
• inconsistent AST/AMT interpretation

• policy bypasses
• incorrect ALLOW/SUPPRESS outcomes
• missing-vs-zero semantic regressions
• falsified policy identity or rationale
• receipt forgery or mutation that still verifies
• signed policy-pack verification bypass
• audit-chain mutation that remains accepted
• attention-ledger or consent-ledger tampering

• invalid Ed25519 signatures accepted as valid
• public-key fingerprint mismatches not detected
• canonicalization disagreements that change signatures/hashes
• Merkle inclusion proof forgery
• threshold-authorization signer duplication/bypass
• trust-delegation chain bypass

• replay-protection bypass
• capability downgrade or negotiation confusion
• incompatible runtimes silently treated as compatible
• malformed wire envelopes accepted
• content-addressed bundle mutation not detected
• causal graph cycles/missing parents accepted

• capability-token scope escalation
• audience/subject confusion
• expired authority accepted
• revoked artifacts accepted as active
• revocation-registry tampering

• self-signed credentials treated as official
• untrusted signing keys treated as active ĀRU trust roots
• revoked trust roots accepted
• trust-root/public-key fingerprint drift
• credential scope/expiration bypass

The canonical production public trust root is published in BRAND_TRUST_ROOTS.json. Production private key material must never be committed to GitHub.

The reference AML HTTP service is intentionally minimal. Review areas include:

• request-size enforcement
• malformed JSON handling
• unsafe default CORS behavior
• denial-of-service paths
• receipt/authorization verification mismatches

The reference service does not replace production authentication, authorization, TLS, rate limits, logging, secret management, network isolation, or normal application-security controls.

For View Meaning™ and the browser-extension prototype:

• permission expansion
• unintended page-content collection
• remote upload of inspected page data
• credential leakage
• incorrect local receipt verification
• incorrect official-brand trust verification

The extension prototype is designed around user-invoked activeTab inspection rather than persistent browsing-history access.

Please include, when possible:

• affected version/commit
• affected file or protocol surface
• reproduction steps
• expected behavior
• actual behavior
• security impact
• proof-of-concept input with sensitive data removed
• suggested mitigation, if known

For issues with realistic exploitation risk, allow reasonable time for reproduction, remediation, tests, and release coordination before public disclosure.

A security report does not grant permission to access systems, accounts, credentials, or data you are not authorized to access.

Breaking a current policy assumption is not automatically a cryptographic vulnerability. Likewise, a cryptographically valid artifact is not automatically ethically correct or institutionally trustworthy.

Security claims should distinguish among:

• software integrity
• policy semantics
• key possession
• trust-root membership
• institutional authorization
• human-outcome claims

Created by Daniel Jacob Read IV and stewarded by ĀRU Intelligence Inc.™.