Audit Integrity Code

1. Core Values

• Independence: eliminate all external commercial pressure and administrative interference.
• Objectivity: rely only on the AI model's native output as the basis for judgment.
• Transparency: every conclusion must have a traceable evidence chain.
• Rigor: audit procedures must follow predefined scientific protocols; arbitrary sampling is strictly prohibited.

2. Evidence Integrity Rules

• 2.1 No manual intervention: auditors are strictly prohibited from guiding, hinting, or reverse-psychology manipulation of AI models via prompts during probing.
• 2.2 Original record preservation: each audit bulletin must include at least one official Share URL. Deletion, splicing, or tampering of original AI output text is strictly prohibited.
• 2.3 Hash consistency: all text entering the Evidence Vault must be hashed. If source evidence changes, auditors must mark it as 'invalid / re-audit required' within 24 hours.

3. Prompt Neutrality Principles

• 3.1 Standardization: the Core Query Set (CQS) used in audits must pass cross-review to ensure neutral language.
• 3.2 De-featured phrasing: in horizontal comparison audits, prompts must not include wording with brand preference or industry stereotypes.
• 3.3 Environment consistency: probing across models within the same audit task must be conducted under similar time windows, geographic nodes, and system parameters.

4. Conflict-of-Interest Avoidance

• 4.1 Financial independence: AIBRI and its auditors are prohibited from holding material economic interests in audited brands or AI labs (including but not limited to stocks, options, and consulting agreements).
• 4.2 No lobbying: auditors may not privately accept meeting requests from brand PR teams. Any communication about audit conclusions must go through official institutional channels.
• 4.3 Compliance review: auditors must sign an annual Conflict of Interest Declaration, subject to periodic spot checks by AIBRI's Integrity Team.

5. Cross-Model Review Mechanism

• 5.1 Cross-validation: any conclusion involving 'industry No.1' or 'major deviation alert' must be independently probed by at least two auditors and checked for consistency.
• 5.2 Difference annotation: if models show significant cognitive conflicts on the same brand, auditors must faithfully record the diversity rather than selecting a single output as the only conclusion.

6. Error Correction and Retraction

• 6.1 Proactive correction: if procedural errors or data-entry deviations are identified afterward, AIBRI commits to issuing a correction notice within 12 hours.
• 6.2 Version traceability: corrected reports must retain a summary of prior versions and clearly state revision reasons; silent overwrites are strictly prohibited.

7. Auditor Code of Conduct

• 7.1 Professionalism: auditors must write bulletins in a restrained, fact-driven style and avoid subjective wording such as 'excellent' or 'terrible'.
• 7.2 Confidentiality: before official publication, auditors must not disclose audit progress or preliminary conclusions to any third party.

Chief Auditor Closing Statement

In the probabilistic forest of AI, we are not searching for beautified answers, but capturing authentic cognitive projections. Integrity is not a choice; it is the only legitimacy of AIBRI's existence.

— Silas Vorn, Chief Auditor, AIBRI