Evidence Vault
The Evidence Vault is the core evidence-preservation system for AIBRI audit conclusions.
In AI auditing, authenticity and immutability are the only foundations of institutional credibility. The Evidence Vault is AIBRI's internal high-security evidence management system. It captures, locks, and archives the raw data stream of every audit session so that all published AIRI indices and benchmark reports retain judicial-grade traceability.
1. Objective
The Evidence Vault exists so that every audit conclusion can be traced back to its original generation context rather than resting on a later summary layer.
It serves both as AIBRI's internal evidence core and as the underlying infrastructure for institutional credibility and audit independence.
2. Operating Logic and Technical Protocol
The Evidence Vault is not a simple list of links but a strict Evidence Chain Protocol (E.C.P.).
A. Raw session anchoring: whenever an auditor starts a probing task, the system automatically generates the official Share URL provided by the AI service provider, such as OpenAI or Anthropic. This URL becomes the first-level evidence and records the model's native probabilistic output under a specific prompt.
B. Digital fingerprint hashing: to guard against future retroactive edits or deletions by AI providers, the Evidence Vault computes a SHA-256 hash for every original text record. Each Audit ID corresponds to a unique digital fingerprint; even if the original link fails, the hash can still prove archival integrity.
C. Multidimensional environment cloning: the system records all environment variables associated with the evidence, including model metadata such as version and system preset, sampling parameters such as temperature, top-p, and max tokens, and geographic node information such as Node: SG-04, Singapore.
D. Visual evidence mirroring: for critical benchmark conclusions, the Evidence Vault automatically captures a full long-form screenshot of the session with an AIBRI audit timestamp watermark as secondary backup evidence.
3. Evidence Tiers and Access Policy
The Evidence Vault is currently in an Internal Restricted stage and applies tiered access according to verification depth.
- L1 Public: official share links released with audit reports and directly verifiable by the public on the AI platform.
- L2 Expert (in preparation): deep evidence packages including full prompt logic and multi-round test variants, intended for future academic partners.
- L3 Core (internal): hash values, node logs, and original mirror captures, restricted to the Chief Auditor for review and compliance examination.
4. Roadmap: Decentralized Evidence Preservation
In future versions, AIBRI plans to migrate Evidence Vault hash records onto a distributed ledger (blockchain) to achieve genuinely third-party-non-interferable evidence preservation.
At that stage, any reader worldwide will be able to verify the authenticity of a report's evidence independently by entering its Audit ID.