Applies to: AITIR Framework Version 2 documentation, schemas, examples, and build/validation scripts

AITIR is a reference architecture and research preview, not a hosted service or production security product. The repository should never contain live credentials, personal telemetry, classified information, protected agency records, or production connector secrets.

Do not open a public issue if a report would expose a credential, unsafe response path, privacy vulnerability, or exploitable implementation detail. Use GitHub private vulnerability reporting. Include:

• affected version, file, schema, or script;
• impact and plausible misuse;
• minimum reproduction without sensitive data;
• whether the issue could bypass decision authority, expiration, idempotency, rollback, or privacy controls;
• suggested mitigation if known.

Do not include real identities or telemetry.

• analytics-to-enforcement authority bypass;
• guard, approval, separation-of-duty, expiration, or replay bypass;
• duplicate, partial, or overbroad connector action;
• evidence or audit destruction;
• feedback poisoning or model/policy release bypass;
• identity-resolution collision or cross-tenant disclosure;
• prompt injection through log, ticket, email, or threat-feed content;
• schema ambiguity that turns advisory evidence into authorization;
• generated artifact that materially misstates its source data;
• repository secret or protected data exposure.

• isolate analytics from production enforcement credentials;
• authenticate and authorize every interface;
• treat all external and log text as untrusted data;
• bind actions to immutable decisions and idempotency keys;
• recheck target preconditions at execution;
• expire decisions and block replay;
• preserve evidence and test rollback;
• quarantine feedback and control model/policy promotion;
• use least privilege, dependency inventory, SBOMs, signed releases, and independent assessment;
• maintain safe degraded modes.

Version 2.0.x documentation receives corrections. Historical v0.1 is retained for research provenance and is not maintained as a safe operational specification.

A report and fix do not certify a deployment. Implementers remain responsible for threat modeling, testing, privacy/legal review, authorization, monitoring, incident response, and secure operations.