Guide 02: Who Governs AI? The Case for On-Chain Accountability

carmenonchain.ai Start Here Series

Guide02

Start Here Series / April 2026

AI Ethics x Blockchain

Who Governs AI?
The Case for
On-Chain
Accountability

“

How blockchain creates enforceable ethical
guardrails for AI systems that current
frameworks cannot.

Where Blockchain and AI
Converge —
and Why It Matters.

Who this is for

AI systems are making decisions that affect millions of people with almost zero accountability. If you have ever wondered who decides how AI behaves, who enforces the rules, and what happens when things go wrong, this guide is for you. It explains how blockchain provides something no other technology currently can: ethical constraints for AI that are transparent, enforceable, and not controlled by any single company.

CONTENTS

What's inside

01

The AI Alignment Problem

Why the most powerful technology on earth has an accountability gap

02

Why Current Frameworks Fall Short

Corporate self-regulation and government regulation compared

03

What Blockchain Brings to AI Ethics

Four capabilities that change the equation

04

Machine-Enforceable Ethics

How consensus mechanisms become ethical constraints

05

AI as Governed AND Governing

The governance loop and why it matters

06

What This Means for You

Why ethical alignment affects everyone

07

Key Terms Reference

Essential concepts in plain language

08

Go Deeper

Research, resources, and next reads

01

SECTION

The AI Alignment Problem

AI systems are becoming more capable every month. They write code, diagnose medical conditions, manage financial portfolios, and generate content indistinguishable from human output. But capability without accountability is dangerous. The alignment problem, in plain language: how do you make sure AI does what we actually want it to do, and how do you prove that it is doing it? Right now, the answer is mostly "trust the company that built it." That is not good enough.

5

Major tech companies control the most powerful AI models globally

0

Enforceable global frameworks for AI ethical compliance

2026

Year autonomous AI agents began managing financial assets independently

WHAT WE CANNOT VERIFY TODAY

Training data

You cannot verify what data was used to train a model

Decision logic

You cannot inspect how a model reaches its outputs

Behavioral constraints

You cannot confirm an AI is following ethical guidelines

Accountability

When AI causes harm, there is no transparent trail of responsibility

02

SECTION

Why Current Frameworks Are Not Working

There are two dominant approaches to AI ethics today: corporate self-regulation and government regulation. Neither is working.

Framework	How It Works	The Problem
Corporate Ethics Boards	Internal review committees, published principles, voluntary commitments	No enforcement mechanism. Companies can dissolve ethics boards, ignore recommendations, and face zero consequences.
Government Regulation	Legislation like the EU AI Act, US state-level bills, agency enforcement	Fragmented across jurisdictions. Years behind the technology. No federal US AI law. Compliance is self-reported.
On-Chain Governance	Rules encoded in smart contracts, enforced automatically, transparent by default, governed by community	Still early. Voter participation challenges, token-weighted voting can concentrate power, technical complexity is a barrier.

Key Insight

On-chain governance is not a perfect solution. But it offers something the other two frameworks fundamentally lack: enforcement that does not depend on the goodwill of the entity being regulated.

03

SECTION

Four Capabilities That Change the Equation

Blockchain does not solve AI ethics by itself. But it provides four capabilities that no other technology currently offers for AI accountability.

Immutable Audit Trails

Every action an AI takes can be recorded on-chain in a way that cannot be altered or deleted.

Example: An AI agent managing a treasury has every transaction, every decision, and every parameter change logged permanently

Smart Contracts as Enforceable Rules

Ethical constraints encoded in smart contracts are not guidelines or best practices. They are code that executes automatically.

Example: A smart contract can prevent an AI from exceeding spending limits, accessing restricted data, or operating outside defined parameters

Transparency by Default

On-chain governance means every proposal, every vote, and every decision is publicly visible and verifiable.

Example: When a DAO votes to change how an AI model operates, the entire decision-making process is recorded and auditable

Decentralized Oversight

No single company, government, or individual controls the rules. Governance is distributed across stakeholders.

Example: The ETHOS Framework proposes using DAOs where developers, regulators, ethicists, and end-users all have a voice in AI oversight

04

SECTION

Machine-Enforceable Ethics

The most compelling idea at the intersection of blockchain and AI ethics is this: what if ethical constraints were not just written down in policy documents but were machine-enforceable? Research from Penn State (Ramljak, 2025) has demonstrated that blockchain consensus mechanisms can function as ethical constraints for AI systems.

Step 1: Define Ethical Constraints

Stakeholders (developers, ethicists, regulators, community) agree on rules for AI behavior. These rules are encoded into smart contracts.

Step 2: Assign Identity

AI agents receive soulbound tokens tied to their identity. These tokens establish credentials, capabilities, and accountability history.

Step 3: Monitor and Enforce

Smart contracts automatically monitor AI behavior against defined constraints. Violations trigger automated responses: alerts, restrictions, or shutdown.

Step 4: Govern and Adapt

Community governance via DAO can update ethical constraints as technology evolves. Every change is proposed, voted on, and recorded transparently on-chain.

RESEARCH CITED

Ramljak 2025 (Penn State/MDPI)

Blockchain consensus as machine-enforceable ethics

ETHOS Framework (arXiv)

Soulbound tokens and DAOs for AI agent governance

VOPPA Framework (MDPI 2025)

AI as governed versus governance tool

05

SECTION

The Governance Loop

The relationship between AI and blockchain governance is not one-directional. AI can be governed by on-chain systems, and AI can also help humans govern more effectively.

	AI as Governed	AI as Governance Tool
What it means	Blockchain constrains AI behavior through smart contracts and enforceable rules	AI helps humans process complex governance proposals, analyze data, and make informed decisions
Example	An AI agent's spending authority is capped by a smart contract. It cannot exceed its budget regardless of what it decides.	MakerDAO uses AI governance tools to summarize 50-page proposals so human voters can understand what they are voting on
Risk	Overly rigid constraints could limit AI's ability to adapt to new situations	Whoever controls the AI that summarizes proposals controls the narrative
Key question	How do you define constraints that are specific enough to enforce but flexible enough to evolve?	Who governs the AI that governs?

PROJECTS TO WATCH

MakerDAO

Using AI governance tools for proposal analysis in the Endgame transition

Vitalik Buterin

Proposed AI stewards where AI helps DAO members process complex decisions while humans retain final authority

NEAR Protocol

Building infrastructure for AI agents with on-chain governance integration

Key Insight

The most important question in AI governance is not 'should AI be governed?' It is 'who governs the AI that governs?' On-chain systems provide the most transparent answer available.

06

SECTION

What This Means for You

You do not need to build smart contracts or run a DAO to care about AI ethical alignment. The decisions being made right now about how AI is governed will shape the technology you use every day.

Who writes the rules?

Today, AI ethics rules are written by the companies that build the models. On-chain governance offers a model where communities, not corporations, define the constraints.

Can the rules be enforced?

Published AI principles are not enforceable. Smart contracts are. The difference between a guideline and a constraint is whether it can be ignored.

Can you verify compliance?

Blockchain transparency means you do not have to take a company's word for it. You can verify that an AI system is operating within its defined constraints.

Can the rules evolve?

Technology moves fast. On-chain governance allows ethical frameworks to be updated through community proposals and votes, not years-long legislative cycles.

07

SECTION

Key Terms Reference

Term	Definition
On-Chain Governance	A system where rules for protocol changes are encoded in smart contracts, allowing stakeholders to vote on and automatically execute updates
Smart Contract	A self-executing agreement stored on the blockchain that automates enforcement based on predefined conditions
DAO	A Decentralized Autonomous Organization governed by smart contracts and community voting rather than corporate hierarchy
AI Alignment	The challenge of ensuring AI systems behave in accordance with human values and intended objectives
Soulbound Token	A non-transferable token tied to a specific identity used to establish credentials, reputation, or accountability
Quorum	The minimum participation threshold required for a governance vote to be considered valid
Timelock	A mandatory waiting period between a governance vote passing and the change being implemented
Quadratic Voting	A voting mechanism where the cost of additional votes increases quadratically, reducing the influence of large token holders
ZKML	Zero-Knowledge Machine Learning, a technique that proves an AI model executed correctly without revealing data or architecture
Plutocracy Risk	The danger that token-weighted voting concentrates governance power among wealthy participants

08

SECTION

Go Deeper

FROM CARMENONCHAIN.AI

Guide 01: Why Blockchain and AI Are Converging	Guide
Guide 03: AI Agents Explained	Coming Soon
Guide 04: Privacy, Decentralization, and AI	Coming Soon
The Weekly Briefing	Newsletter · carmenonchain.ai/newsletter

RESEARCH CITED

Ramljak 2025, Penn State/MDPI	Blockchain consensus as machine-enforceable ethics
ETHOS Framework (arXiv)	Soulbound tokens and DAOs for AI agent governance
VOPPA Framework (MDPI 2025)	AI as governed versus governance tool
Saesen et al. 2026 (TU Dortmund/JIT)	On-chain DAO governance and token performance

EXTERNAL RESOURCES

Chainlink: On-Chain Governance Explained	chain.link
Stanford HAI	hai.stanford.edu · AI policy and ethics research
Vitalik Buterin	vitalik.eth.limo · AI stewards proposal

This guide is part of the Start Here series on carmenonchain.ai

Carmen Onchain | @carmen_onchain | carmenonchain.ai