Files
hips/HIPs/hip-0230-ai-transparency-explainability.md
zeekay ad5a010f79 Three words for what a proposal is, and a generator that refuses
The same fact lived in nine hand-kept places and they disagreed about 58
proposals. This makes the file the only authority and everything else a
projection of it.

ONE VOCABULARY, in vocabulary.json, read by the lint, the generator and the
doc site. Three words, each answering "can I rely on this?" and nothing else:

    Draft    nothing implements it yet, or its own text says the work is ahead
    Final    the thing it specifies exists in the code
    Living   governance and process: amended in place, never finished

Eight spellings folded in. `Active` was the one doing real damage: on four Meta
and Process HIPs it meant living, on 48 others it meant shipped, and one word
doing two jobs is why nothing agreed. `Review` and `Last Call` described a
queue rather than a document, and a proposal under review is still one you
cannot build against.

STATUS IS EARNED, not declared. Every Standards Track proposal was checked
against the estate: 235 have their implementation, and eight do not and say so.
HIP-0095 carries a twelve-week plan ending at "mainnet launch" and has no code.
HIP-0525 specifies an all-machine meeting while apps/meet refuses a machine a
seat. HIP-0078 names a prover at a "planned location" that does not exist.
HIP-0114 is the cog.* procedure set, and that procedure set returns nothing.
HIP-0003, HIP-0102, HIP-0305 and HIP-0901 likewise: looked, found nothing.

SIXTEEN TOMBSTONES DELETED rather than statused. A dead document kept alive in
the index is one concept wearing two names. Each successor exists and carries
the text, and the 90 references that pointed at a tombstone now point at it --
whole markdown links at once, so link text cannot outlive its target. Three of
the sixteen were pure compatibility: HIP-0028 specified a migration off a
Bitnami chart, HIP-0029 dual-engine MySQL, HIP-0030 a schema-deprecation
policy. We ship no backwards compatibility, so we do not specify it.

Nine links were already broken and had been since the corpus was renumbered:
short-form ./hip-5.md paths that never existed. Two named the wrong proposal --
one cited "HIP-1: Hanzo Multimodal Models" where multimodal is HIP-0003.

THE GENERATOR REFUSES more often than it writes, because a short table looks
exactly as authoritative as a complete one. Rows must equal files; a shrinking
index must say by how much; a section whose input is missing is an error, not
an omission; nothing is written until all of it validates. Its self-test breaks
the corpus eleven ways and requires each to refuse AND leave README byte-
identical -- a guard that reports after half-writing has protected nothing.
The lint's second implementation of the same comparison is gone.

The site rendered 52 proposals with a fallback grey badge and counted them
toward none of its three stat tiles, because docs/lib/source.ts declared a
status union by hand: it admitted a `Stagnant` no file has ever carried and
omitted the one 52 files did. Five components each kept their own copy of the
colours. All six now read vocabulary.json.
2026-08-28 18:05:41 -07:00

9.5 KiB

hip, title, tags, description, author, discussions-to, status, type, created, requires
hip title tags description author discussions-to status type created requires
230 AI Transparency & Explainability
ai-ethics
transparency
explainability
governance
Framework for ensuring transparency and explainability of Hanzo AI systems. Hanzo AI Team (@hanzoai) https://github.com/hanzoai/hips/discussions Living Meta 2025-12-17 HIP-0200, HIP-0201

HIP-0230: AI Transparency & Explainability

Abstract

This HIP establishes the transparency and explainability framework for Hanzo AI systems. It defines requirements for communicating AI capabilities and limitations, providing explanations for AI outputs, and ensuring stakeholders can understand and audit AI behavior.

Transparency Framework

Transparency Principles

  1. Clarity: Information is understandable to intended audience
  2. Accessibility: Information is easy to find and access
  3. Completeness: Material information is disclosed
  4. Accuracy: Information is correct and up-to-date
  5. Proportionality: Level of disclosure matches risk level

Transparency Levels

Level Audience Depth Examples
Public General users High-level Product pages, blog posts
User Active users Functional In-product disclosures
Developer API users Technical API documentation
Auditor Reviewers Detailed Model cards, audit reports
Regulator Authorities Comprehensive Regulatory filings

AI Disclosure Requirements

System-Level Disclosure

AI Presence Disclosure

Requirement: Users must know when they're interacting with AI.

Context Disclosure Method
Chat interface Clear "AI" label
Voice interface Audio disclosure
Generated content Watermark/label
Automated decisions Explicit notice

Capability Disclosure

Requirement: Communicate what the AI can and cannot do.

Element Disclosure
Intended use Primary use cases
Limitations Known weaknesses
Not suitable for Inappropriate uses
Accuracy expectations Performance levels

Model-Level Disclosure

Model Card Requirements

Every model must have a model card containing:

Section Required Contents
Model details Name, version, type, architecture
Intended use Primary uses, users, out-of-scope uses
Training data Data sources, composition, limitations
Performance Benchmark results, evaluation methodology
Limitations Known limitations, failure modes
Ethical considerations Bias, risks, mitigations

Training Data Disclosure

Element Disclosure Level
Data sources Named sources where possible
Data composition Categories, proportions
Data processing Filtering, cleaning methods
Data limitations Known gaps, biases

Output-Level Disclosure

Generated Content Labeling

Content Type Labeling Requirement
Text AI-generated indicator
Images Watermark + metadata
Audio Audio watermark + metadata
Video Visual indicator + metadata

Confidence Disclosure

When appropriate, indicate confidence:

  • Uncertainty indicators for factual claims
  • Confidence scores for classifications
  • Probability ranges for predictions

Explainability Framework

Explainability Levels

Level Description Audience
Functional What the AI does End users
Behavioral Why the AI responded this way Users, developers
Technical How the AI works internally Experts, auditors

Explanation Types

Contrastive Explanations

"Why X instead of Y?"

Use Case Approach
Classification Why this class, not another
Generation Why this response, not alternative
Recommendation Why this item, not others

Counterfactual Explanations

"What would change the outcome?"

Use Case Approach
Decisions What input changes would change result
Refusals What would make request acceptable

Feature Attribution

"What influenced this output?"

Use Case Approach
Text Highlight influential words/phrases
Images Show attention/saliency maps
Structured Show feature importance

Explanation Implementation

User-Facing Explanations

Context Explanation Type
Content refusal Reason for refusal
Uncertain response Confidence indication
Sourced claims Citation/reference
Recommendations Relevance factors

Developer-Facing Explanations

API Feature Purpose
Logprobs Token probability information
Confidence scores Output certainty
Reasoning traces Chain-of-thought (where applicable)

Audit-Facing Explanations

Artifact Contents
Training logs Training process documentation
Evaluation results Detailed benchmark performance
Decision logs Sample decision explanations
Attention analysis Model attention patterns

Documentation Standards

Public Documentation

Document Contents Update Frequency
Product page Capabilities, use cases As features change
Help center How to use, limitations Continuous
Blog/announcements Major updates, changes As needed
Research papers Technical details On publication

Technical Documentation

Document Contents Audience
API documentation Endpoints, parameters, examples Developers
Model card Model details, performance, limitations All
System card System-level information Auditors
Safety documentation Safety measures, testing Regulators

Internal Documentation

Document Contents Access
Training documentation Data, process, decisions Internal + audit
Risk assessments Identified risks, mitigations Internal
Incident reports Safety incidents, responses Internal + regulators

Audit & Verification

Internal Audit

Audit Type Frequency Scope
Documentation review Monthly Accuracy, completeness
Disclosure compliance Quarterly All disclosure requirements
Explanation quality Quarterly User understanding

External Audit

Audit Type Frequency Auditor
Model audit Annual Third-party ML experts
Documentation audit Annual Compliance experts
User understanding study Biennial Research partners

Verification Methods

Method Purpose
User surveys Verify understanding of disclosures
A/B testing Test explanation effectiveness
Expert review Technical accuracy verification
Red team Attempt to find undisclosed capabilities

Special Contexts

High-Stakes Decisions

When AI influences significant decisions:

Requirement Implementation
Explicit AI role Clear statement of AI's role
Human oversight Indication of human review
Appeal process How to contest decisions
Detailed explanation Factors that influenced outcome

Synthetic Content

For AI-generated content:

Requirement Implementation
Labeling Clear AI-generated indicator
Watermarking Technical watermark in media
Provenance Metadata about creation
Detection tools Tools to verify AI origin

Research & Development

During model development:

Requirement Implementation
Internal documentation Track decisions, data, methods
Reproducibility Enable result reproduction
Version control Track model versions
Change documentation Document capability changes

Governance

Transparency Oversight

Role Responsibility
Transparency Lead Day-to-day compliance
Communications Team Public-facing content
Legal Team Regulatory compliance
ESG Committee Policy oversight

Review Process

Activity Frequency Participants
Disclosure review Monthly Transparency Lead
Documentation audit Quarterly Cross-functional
Policy review Annual ESG Committee

Escalation

Issue Escalation Path
Disclosure gap Transparency Lead → Product Lead
Misleading content Communications → Legal → ESG Committee
Regulatory concern Legal → Board
  • HIP-200: Responsible AI Principles
  • HIP-201: Model Risk Management
  • HIP-210: Safety Evaluation Framework
  • HIP-220: Bias Detection & Mitigation
  • HIP-240: AI Incident Response
  • HIP-250: Sustainability Standards Alignment

Changelog

Version Date Changes
1.0 2025-12-17 Initial draft

Copyright and related rights waived via CC0.