Skip to content

Repository files navigation

ATI — Auditable Trust Infrastructure

The category for proving what a regulated AI system did, without keeping the data that proves it.

Category Status PII Regulatory License


"Trust should not be a promise. It should be a transaction."

Auditable Trust Infrastructure (ATI) is an architectural category — a Layer-0 standard for how AI, data, and trust interact inside regulated institutions. It exists to resolve a conflict that policy and process cannot: the law requires you to retain the evidence of a decision for years, and simultaneously requires you to erase the personal data behind it on request.

ATI answers the question that breaks conventional compliance:

How do you prove you made the right decision if you deleted the data?

This repository is the public canon of the category: the narrative that defines it, the frameworks that implement it, the products built on it, and the cryptographic contracts that make its claims verifiable rather than rhetorical.

👉 New here? Start with the category narrative or jump to the Master Index.


Why a category, not a product

Most "compliance AI" is probabilistic scoring with a confidence number stapled on. ATI is a different posture: trust by physics, not trust by policy. You do not promise an auditor you behaved; you hand them a cryptographic proof that you could not have behaved otherwise.

Conventional approach The ATI posture
Logic Probabilistic ("maybe fraud") Deterministic ("policy violated at line 42")
Data Leaky (needs PII at rest) Capsulated (PII-free state objects)
Audit "Explainable" (post-hoc rationalization) Verifiable (cryptographic proof of causality)
Erasure Conflicts with retention Crypto-shredding (retain ciphertext, destroy the key)
Trust "We promise not to look" "We cannot look"

The three pillars of Trust by Physics

graph TD
    P1[1 · Zero-Persistence Runtime]
    P2[2 · Immutable WORM Ledger]
    P3[3 · Crypto-Shredding Layer]
    P1 -->|generates proofs, not data| P2
    P2 -->|anchored by destroyable keys| P3
Loading
  1. Zero-Persistence — sensitive data lives in volatile memory only for the milliseconds a decision takes. No disk write. When the transaction closes, the data evaporates.
  2. Immutable WORM Ledger — the system stores proofs, not payloads: the cryptographic hash of every decision, logic path, and outcome, write-once and tamper-evident.
  3. Crypto-Shredding — "forgetting" a subject means destroying the encryption key for their records. The ciphertext stays (retention satisfied); without the key it is mathematical entropy (erasure satisfied).

Full treatment in category/02 — Trust by Physics.


Whitepapers

ATI Category Whitepaper

The ATI whitepaper defines Auditable Trust Infrastructure as a category for verifiable AI evidence in regulated environments.

  • Markdown source: whitepaper/ati/FoundLab_ATI_Whitepaper_v1.1.md
  • PDF: whitepaper/ati/FoundLab_ATI_Whitepaper_v1.1_A4.pdf
  • Source HTML/CSS: whitepaper/ati/source/

F2F-RaaT Framework Whitepaper

F2F-RaaT is a normative execution framework within ATI. It is not the category itself.

  • Specification: whitepaper/f2f-raat/F2F-RAAT_WHITEPAPER_v1.1.md
  • Diagrams: whitepaper/f2f-raat/diagrams/

How this repository is organized

ATI/
├── category/                  # THE CREATION OF THE CATEGORY — start here
│   ├── 01 · The Regulatory Paradox
│   ├── 02 · Trust by Physics (the three pillars)
│   ├── 03 · Cognitive Auditability (taming generative AI)
│   └── 04 · Strategic Impact
│
├── modules/                   # THE BUILDING BLOCKS
│   ├── frameworks/            #   the open specifications
│   │   ├── f2f-raat/          #     From Fact to Feedback — deterministic execution engine
│   │   ├── spezzatura/        #     the T² reputation model (the math)
│   │   ├── state-capsule/     #     the PII-free boundary object (the memory)
│   │   ├── burn-engine/       #     deterministic authorization (the actuator)
│   │   ├── veritas/           #     immutable audit trail & proofs (the truth)
│   │   └── threat-model/      #     adversarial analysis & mitigations
│   └── products/              #   what gets built on the frameworks
│       └── rex-guard/         #     cryptographic governance layer for generative AI
│
├── reference-architecture/    # public, sanitized reference SAD for regulated GenAI
├── whitepaper/                # the deep-dive theory
│   ├── ati/                   #   ATI category whitepaper (PDF & sources)
│   └── f2f-raat/              #   F2F-RAaT framework whitepaper
├── governance/                # policy signing, break-glass control
├── compliance/                # regulatory mapping + test vectors
├── interface-contracts/       # OpenAPI + canonical Fact schemas
├── cookbooks/                 # real-world scenario walkthroughs
└── GLOSSARY.md                # canonical vocabulary of the category

Positioning & honesty

ATI is described here as architected for and aligned with the regulations it addresses (LGPD, GDPR, BCB / CMN cybersecurity norms, EU AI Act). Certification and production status are tracked per deployment, not asserted blanket-wide. The frameworks in this repo are normative specifications; the metrics that appear in the narrative come from controlled validation and are flagged as illustrative where they do.

This is deliberate. A category whose central claim is verifiable trust cannot afford an unverifiable sentence in its own README.


License

Authored by FoundLabAuditable Trust Infrastructure. Licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International. See LICENSE.

This repository publishes category-level documentation and reference architecture materials. It is not an open-source software implementation and does not grant rights to use FoundLab private systems, trademarks, customer deployments, or NDA materials.

don't trust, verify.

About

Category definition, reference architecture, and public technical primitives for Auditable Trust Infrastructure — Trust by Physics for regulated AI execution.

Resources

Code of conduct

Contributing

Security policy

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Used by

Contributors

Languages