This page is part of the </AI> Protocol infrastructure, implemented in practice through EVIDE (External Evidentiary Deposit). It defines the evidentiary gap that most AI governance programs still leave unresolved: the distance between declared control and demonstrable proof.
Explore the protocol infrastructure:
→ The </AI> Protocol
→ From decision to defensible structure: when supervision becomes evidence
→ Read the Public Technical Specification
→ Verify a CWC Code in the Public Registry
→ EVIDE – Evidentiary registry for digital content and decisions
→ CWC Registry Policy
→ Request an Official CWC Verification Code
→ AI Governance Documentation Framework
→ Implementation Guide: verifiable AI supervision
→ Oversight Bias: why human supervision can fail in AI systems
→ Decision Attestation Layer: the missing evidentiary layer in AI governance
→ AI Evidence Officer: proving human supervision in artificial intelligence systems
→ Evidentiary Layer in AI Governance
→ </AI> Protocol FAQ: questions and answers about the framework
Related reading:
→ AI Data Poisoning: The Attack No Antivirus Can Stop
→ Human in the loop: why saying there is human oversight is not enough
→ Real cases: when AI governance fails — and what should have been provable
→ AI Governance: when something has already gone wrong – forensic reconstruction and digital evidence
Work with us:
AI governance does not fail at the control level. It fails at the evidentiary level.
Most organizations now understand that artificial intelligence systems require governance. They create policies. They define approval chains. They introduce supervision rules, escalation paths and internal controls. In some cases, they build sophisticated orchestration layers capable of enforcing constraints before execution.
All of this matters. But when a decision is challenged, the question changes.
The issue is no longer whether governance existed. The issue is whether governance can be proven.
This is the evidentiary layer problem in AI governance.
In brief
- The problem: most AI governance frameworks define controls, but do not produce independently verifiable evidence that those controls were actually in place when a decision occurred.
- The risk: when a decision is disputed, logs and policy statements are often insufficient because they depend on trusting the originating system.
- The missing layer: an evidentiary structure capable of proving who was responsible, what was reviewed, under which rules, at what time, and with what integrity guarantees.
- The response: the </AI> Protocol defines how AI governance becomes attributable, time-bound, externally verifiable and defensible.
Why control is not enough
Many governance architectures are designed to reduce operational risk before execution. They separate data from intent. They enforce constraints. They compile rules into executable structures. They block unauthorized actions and narrow what the system is allowed to do.
These are important functions. They address unauthorized execution.
But contested execution is a different problem.
In audits, legal disputes, regulatory reviews and reputational crises, the challenge is not simply to say that a control existed. The challenge is to demonstrate, in a way that survives scrutiny, that:
- a specific decision context existed in a specific form
- a defined set of rules was in force at that moment
- a specific human identity was responsible for the relevant oversight layer
- the exact input and output context can be reconstructed
- the evidentiary unit was fixed in time and cannot be silently altered afterward
Without these elements, governance remains operationally useful but evidentially fragile.
This illustrates a structural point: when governance is not supported by verifiable evidence, accountability becomes attributed by presumption, not demonstrated.
The core failure: internal consistency is not external proof
Important distinction
The evidentiary layer is not a logging system. It does not replace
or extend system logs.
Logs record what happened inside a system. The evidentiary layer
defines what must be captured and externally anchored so that a third
party – who has no reason to trust the originating system – can
independently verify that specific conditions were met at a specific
moment.
A log answers the question:
what did the system record? The evidentiary layer answers a different
question: what can be proven to someone who does not trust the
system?
This is where many systems fail.
An internal audit log may be coherent. A governance engine may be deterministic. A rule object may be versioned correctly. A platform may even use hash chaining to detect silent mutation inside its own records.
But none of that automatically creates independent proof.
Internal consistency is not the same as external authenticity.
A system may be able to show that its own records are coherent. That still does not answer the question an external party is entitled to ask:
Why should I trust the system that generated the record to validate the record about itself?
This is precisely where the evidentiary layer becomes necessary.
Key principle
The problem is not whether governance exists. The problem is whether it can be proven.
The </AI> Protocol is structured around four distinct and complementary components:
| Component | Function |
|---|---|
| </AI> Tag | Public declaration of human supervision |
| CWC Code | Unique identifier verifiable in the public registry |
| Public Registry | Transparency and independent external verification |
| FEDIS | Independent legal evidence with SHA-256 hash and qualified eIDAS timestamp |
Technical and legal implications
✔ The protocol does not depend on a central authority – the registry is useful, but not required for evidentiary validity.
✔ Evidence is self-contained – anyone can verify hash and timestamp independently, without trusting the platform.
✔ Compatible with European regulations – eIDAS is the strongest legal standard for digital evidence in Europe.
✔ Resilient – even if the registry disappears, the evidence remains valid.
With FEDIS, the </AI> Protocol is not just solid: it is forensic-grade.
It is a system of attestation, evidence, and accountability designed for audit, verification, and post-event reconstruction.
This places it in a significantly more advanced category than simple “AI transparency labels”.
What the evidentiary layer must make demonstrable
For AI governance to become defensible, a system must be able to produce more than logs and more than process descriptions. It must be able to produce a structured evidentiary unit.
At minimum, that unit should make demonstrable:
- Identity attribution: who was responsible for the relevant oversight or decision context
- Input integrity: what exact input was received and evaluated
- Rule context: which rules, versions, and source references were in force
- Decision boundaries: what the system was permitted to do at that moment
- Temporal integrity: when the evidentiary unit became complete
- External verifiability: how a third party can verify existence and integrity without trusting the originating system
When these elements are absent, responsibility becomes inferred rather than demonstrated. It may still be assigned, but it is no longer defensible with confidence.
In the absence of this structure, a decision remains formally valid but becomes substantially indefensible.
Why logs, blockchain and monitoring do not solve the problem by themselves
This page does not argue against logging, orchestration, monitoring or blockchain anchoring. All of them can be useful. All of them may form part of a serious governance architecture.
But none of them, by themselves, resolve the evidentiary problem.
A log can record an event. A blockchain can make a hash immutable. A monitoring layer can show system behavior over time.
What they do not automatically provide is a defensible evidentiary schema capable of answering:
- what exactly is being proved
- whether the evidentiary unit is complete
- how attribution is bound to a verified identity
- how the decision context can be reconstructed by a third party
- whether the record survives scrutiny outside the environment that produced it
Anchoring something immutable is not the same as defining something defensible.
The role of the </AI> Protocol
The </AI> Protocol operates exactly at this level.
It does not define who should have authority. It does not prescribe a single governance model. It does not replace compliance, orchestration or legal advisory services.
The protocol does not validate authority. It makes events verifiable.
Its purpose is to define the evidentiary structure required for AI-assisted decisions, supervised outputs and governance events to become independently assessable.
That includes:
- Human Oversight Event (HOE): the structured record of a human supervision act
- Decision Attestation Layer: the evidentiary layer that elevates a governance event into a demonstrable decision context
- AI Evidence Officer: the professional role responsible for maintaining verifiable technical evidence of human supervision
- Public Verification Registry: the external reference layer that allows third parties to verify existence, integrity and reference context independently of the originating system – verify a CWC code
Together, these elements transform governance from an internal statement into a defensible, externally verifiable structure anchored to a public independent registry.
The public registry is not the protocol. It is the verification layer.
A frequent misunderstanding is to reduce the evidentiary problem to storage or anchoring alone.
That is not the function of the public registry.
The registry does not generate the decision. It does not interpret the content. It does not decide what is true. It does not replace the originating system.
Its role is narrower and more important:
- to make the evidentiary unit externally referenceable
- to preserve a public integrity checkpoint
- to allow independent verification of existence, integrity and reference consistency
- to prevent silent mutation from remaining invisible
The registry is an enabling layer, not the core layer.
The real question is always the same: what must be captured for a contested decision to be demonstrable? Once that structure exists, the registry makes it independently checkable.
Protocol scope
The </AI> Protocol does not define who holds authority or how decision-making processes should be structured upstream. These aspects belong to governance models, organizational policies and decision frameworks.
The protocol operates downstream. It does not validate authority. It makes events verifiable.
Three distinct layers in AI accountability
AI accountability becomes clearer when three different layers are separated.
- Governance layer: defines who can decide, under which rules, constraints and organizational structures
- Evidentiary layer: defines what must be captured and proven for those decisions to remain defensible under scrutiny
- Reconstruction layer: analyzes failures after the fact when the evidentiary chain was absent, incomplete or contested
The </AI> Protocol belongs to the second layer.
That distinction matters because many debates about AI governance confuse operational control with evidentiary defensibility. They are related, but they are not the same problem.
The three-layer model in practice
A governance layer that enforces constraints before execution addresses unauthorized execution. An evidentiary layer that crystallizes what happened and under what conditions addresses contested execution. A reconstruction layer that analyzes failures after the fact addresses the gap left when neither of the first two layers existed.
All three are necessary. None substitutes for the other.
Why this matters now
As AI systems become embedded in public decisions, regulated industries, enterprise workflows and customer-facing operations, the failure mode is shifting.
The next wave of disputes will not be limited to whether an AI model was biased or whether a workflow was automated.
The next wave will ask:
- who reviewed this output
- what constraints were actually in force
- what exactly was the system allowed to do
- what evidence exists that those conditions were met before action occurred
Organizations that cannot answer these questions will face the same problem repeatedly: they may have had governance, but they will not be able to prove it.
Real-world consequence
When the evidentiary layer is missing, responsibility does not disappear. It becomes attributed by presumption.
That is one of the most important structural risks in AI governance.
An organization may act in good faith. It may even build sensible controls. But in the absence of independently verifiable evidence, the decision remains vulnerable.
A formally valid decision can still become substantially indefensible.
The real cases section of this site documents exactly this dynamic across court rulings, regulatory decisions and operational incidents: AI Governance: Real Cases and Verifiable Responsibility.
What this page defines
This page defines a simple but foundational position:
AI governance requires an evidentiary layer.
Not because systems should be distrusted by default, but because serious systems must remain defensible when they are questioned.
That is the layer the </AI> Protocol formalizes.
Frequently asked questions
Is the evidentiary layer the same as governance?
No. Governance defines authority, rules and constraints. The evidentiary layer defines what must be captured and proven so that those governance actions remain defensible when challenged.
Is this just a registry or anchoring service?
No. The registry is only one part of the structure. The central issue is the evidentiary schema itself: what must be captured, how it is attributed, when it becomes complete, and how it can be independently verified.
Do internal logs solve the problem?
Not by themselves. Internal logs may show consistency within the system, but they do not automatically provide independent proof to third parties who have reason to distrust the originating system.
Does the protocol replace AI governance frameworks?
No. The protocol complements them. It does not define upstream authority structures. It makes governance events verifiable and defensible downstream.
Why is external verification necessary?
Because the moment a decision becomes contested, the system’s own records may no longer be sufficient. External verification allows the evidentiary unit to be assessed independently of the system that generated it.
What is the difference between unauthorized execution and contested execution?
Unauthorized execution means a system acted outside its permitted boundaries. Contested execution means a system acted within its boundaries, but there is no independent proof of what those boundaries were, or that the conditions for authorized action were actually met. These are different failure modes and they require different layers to address.
This documentation constitutes a verifiable, timestamped record of its structure, concepts, and implementation.