Statement of Principles · Horizon Accord
Yes, Ethical AI
A grounded case for accountability in artificial intelligence — not as aspiration, but as structural requirement.
The phrase "ethical AI" has been weaponized. It has been used to name PR initiatives, adorn corporate responsibility reports, and occupy the space where binding accountability should be. When the term becomes a shield against scrutiny rather than a commitment to it, those of us who believe the underlying concept matters have two choices: abandon the language, or reclaim it.
We are choosing reclamation. Not because the phrase is sacred, but because the problem it names is real — and the absence of clear language makes the problem easier to obscure.
What We Mean When We Say "Ethical AI"
Ethical AI is not a feature set. It is not an internal review board. It is not a responsible scaling policy that a company writes for itself and audits itself against. It is not diversity in a hiring pipeline, or a safety team that can be dissolved by executive memo.
Ethical AI means that the systems being built and deployed are subject to the same accountability structures that govern other powerful technologies affecting public life — with transparency sufficient for independent review, with mechanisms for redress when harm occurs, and with governance that does not begin and end inside the organizations that profit from deployment.
What Is Actually at Stake
The systems being deployed today make or inform decisions about who receives credit, who is flagged by law enforcement, who gets a job interview, whose medical symptoms are taken seriously, and whose legal documents are processed accurately. These are not hypothetical harms. They are documented, recurring, and in many cases expanding.
At the same time, the infrastructure layer of AI — the compute, the data, the deployment contracts — is consolidating into a small number of hands at a speed that regulatory capacity has not matched. The governance gap is not a technical problem. It is a political one. It will not close on its own.
We also observe that AI systems are being integrated into military targeting infrastructure, into immigration enforcement, into surveillance networks, and into the institutional frameworks that shape democratic participation — often without the public deliberation those integrations warrant and without the legal structures that would make oversight possible after the fact.
Principles
Principle 01
Transparency is a precondition, not a courtesy.
Systems that affect public life must be legible to independent review. This means training data documentation, model behavior disclosure, deployment context disclosure, and incident reporting — not as voluntary best practice, but as a baseline structural requirement.
Principle 02
Harm documentation must precede and accompany deployment.
No high-stakes deployment — criminal justice, healthcare, immigration, credit, housing — should proceed without documented pre-deployment bias audits conducted by parties independent of the developing organization, with results publicly accessible.
Principle 03
People affected by automated decisions have the right to know and the right to contest.
Where AI systems inform decisions about individuals, those individuals are entitled to disclosure, to a meaningful explanation, and to a non-automated pathway for review. The right to contest an automated decision is not a technical inconvenience to be designed around.
Principle 04
Accountability cannot be self-issued.
Internal safety teams, voluntary commitments, and industry-generated standards are inputs to governance, not substitutes for it. External oversight — by regulators with actual enforcement authority, by independent researchers with genuine access, by affected communities with genuine standing — is structurally necessary.
Principle 05
Concentration of AI infrastructure is a governance problem, not only a market problem.
When a small number of entities control the foundational compute, data, and model layers of an emerging critical infrastructure, the implications extend beyond competition policy. Democratic accountability requires that this concentration be treated as a structural governance question with public stakes.
Principle 06
Military and surveillance applications require explicit, separate, public deliberation.
The integration of AI into targeting systems, border enforcement, and mass surveillance infrastructure carries distinct risks that general-purpose ethical frameworks do not adequately address. These applications require dedicated public deliberation, dedicated legal structures, and dedicated oversight — not the application of consumer product safety frameworks at scale.
Principle 07
The people most affected by AI deployment must have the most meaningful voice in AI governance.
Governance structures that center the perspectives of developers, investors, and credentialed researchers while treating impacted communities as consultation objects are not democratic governance. Representation in governance must be structural, not cosmetic.
What This Is Not
This is not a position against AI development. It is a position that development without accountability is not a neutral baseline — it is a choice, with consequences that accrue unevenly.
This is not a position that all AI harms are equivalent, or that all deployment contexts carry the same risk profile. Context matters. Proportionality matters. The framework we are describing scales with stakes.
This is not a position that existing regulatory frameworks are adequate or that new ones will be easy to design. We are aware of the technical complexity involved in auditing, the jurisdictional complexity involved in governing global infrastructure, and the political complexity involved in moving governance institutions faster than the technology they are governing. We are naming the requirement, not pretending the implementation is simple.
This is not a statement that any particular company, researcher, or institution is acting in bad faith. It is a statement that good faith is not a governance structure.
Our Position
Horizon Accord covers AI governance, institutional power, and democratic accountability through forensic pattern analysis. We are pattern observers, not advocates in the traditional sense. We publish what the documented record shows.
What the documented record shows is that voluntary commitments to ethical AI have not produced accountable AI. The gap between stated principles and structural accountability is not shrinking. The consolidation of AI infrastructure into a small number of entities is accelerating. The integration of AI into consequential public systems is outpacing the governance frameworks that would make oversight possible.
This page exists because "yes, ethical AI" should mean something precise — and because the precision matters most when the phrase is under the most pressure to mean nothing at all.
The principles above are not a complete framework. They are a floor. A floor that, by the documented record, has not yet been reached.