ATAL – AI Traceability & Accountability Ledger

Public Review Draft v0.9 – AI Traceability, Auditability, and Autonomy Governance

View the Project on GitHub Elytra-Security/atal-standard

Responsibility of Implementers

Public Review Draft v0.9


1. Purpose

This document explains the responsibilities of organisations and developers who implement, integrate, or operate ATAL-compliant systems.

Responsibilities apply across the entire ATAL specification, which is structured into nine Parts (I–IX).

Implementers of ATAL are responsible for ensuring that ATAL-based systems operate as non-intervening systems of record. ATAL implementations must not enforce policies, block execution, modify AI behavior, or introduce control logic into AI systems.Implementers remain solely responsible for legal, regulatory, and ethical compliance arising from AI system deployment. ATAL provides technical accountability mechanisms but does not transfer or dilute responsibility.


2. Core Obligations

Complete Instrumentation

Implementers must ensure that all required ATAL records are generated, including Decision Trails, SML entries, EIT signals, and CAG linkages.

Selective instrumentation is non-compliant.

Preserve Evidence Integrity

All ATAL evidence must be cryptographically signed, tamper-evident, immutable, and traceable.

No modification, deletion, or concealment is permitted.

Uphold the 0th Law

Any system capable of autonomous action must be governed by an external, independent accountability layer capable of observing, restricting, or terminating the AI’s actions.

This requirement cannot be bypassed.


3. Human-Initiated Interactions

Implementers must:

Human-triggered actions must produce Decision Trails in accordance with the specification.


4. Autonomous System Responsibilities

Systems capable of self-initiated or escalatory behaviour require:

Autonomy without these elements is non-compliant.


5. Safety and Oversight Requirements

Implementers must provide:

Active Safety Kernel

A functioning, independent layer that can:

AI actions in real time.

Verified Kill-Switch Pathways

Kill-switch actions must be immediate, logged, and irreversible by the AI system.

Escalation Controls

High-risk or emergent behaviours must trigger mandatory oversight.


6. Lineage and Causality

Implementers must ensure:

Breaks in lineage undermine ATAL compliance and must be treated as critical failures.


7. Cryptographic and Key Management Responsibilities

Implementers are responsible for:

Compromised keys invalidate ATAL evidence.


8. Operational Responsibilities

Implementers must deploy:

They must also ensure:


9. Prohibited Practices

The following are strictly prohibited:

Any violation constitutes a critical breach.


10. Compliance Responsibilities

Implementers must:


11. Enforcement

Non-compliance may result in:

ATAL evidence may be used in legal or regulatory investigations.


12. Status of This Document

This document applies to ATAL v0.9 and will be updated after the public review period.


End of document.