PLAIN-LANGUAGE ARTICLE · 30 OCTOBER 2026 · PUBLIC EXPLANATION
Who Writes the Rules, Who Carries the Blame?
The research explained in everyday language
When a company builds an AI system, people outside the company may advise it about right and wrong. But advice is not the same as deciding what the system will actually do. The person or organization that gives advice may not be the one that approves, installs, changes, or can reverse the rule. That difference is what this research investigates.
Publication status: This article explains the separate, author-approved Final v1.0 research paper. It is an informal, mutable reader guide, not an additional locked edition. The underlying research is AI-assisted, unvalidated, not independently peer-reviewed, and not a proven AI safety mechanism.
Think about a rule at work
Imagine a factory brings in an outside safety adviser. The adviser recommends a change. Management approves a policy, maintenance or engineering puts it into practice, and employees must follow it. Later, something goes wrong. Who should answer for the decision?
You would want to know who made the recommendation, who actually approved the change, what was installed, when it was installed, who could stop or reverse it, and who was responsible for correcting the harm. Blaming the adviser simply because the adviser was in the room would leave important questions unanswered.
This is an illustration, not a case studied in the paper. AI governance can have similar splits in authority, even though the technology and legal responsibilities are different.
Now apply that question to artificial intelligence
AI companies can consult technical specialists, ethicists, religious communities, civil-society organizations and governments. Consulting them may be thoughtful, useful, or even necessary. But a public announcement saying “we consulted experts” does not tell us whether those experts controlled any final decision.
An adviser might recommend one thing and a developer might implement another. A policy might change between a test version and the system people actually use. An organization might talk publicly about its advisers while retaining every power to approve, modify, or withdraw the rule.
The paper asks us to separate these different responsibilities rather than treat them as the same thing. It does not claim that religious advisers or AI companies have deliberately shifted blame.
What the paper proposes
The research proposes an authority–accountability provenance crosswalk. That is a technical way of saying: a record that connects a specific piece of advice to a specific decision, a particular deployed AI version, an actual outcome, and the person or institution that could provide a remedy.
In ordinary terms, the record would try to answer these questions: Who suggested the rule? Who approved it? Who wrote or deployed the implementation? Which version was affected? Who could change or override it? Who had the power to remedy harm? What evidence connects those steps?
The proposed method is deliberately cautious about missing records. “We do not know” must not quietly become “someone was hiding something.” A lack of documentation is a research obstacle, not proof of deception or wrongdoing.
What happened when the idea was checked?
This is where the paper reports an important unfavorable result. The research examined an exploratory set of 11 case records to see whether public documents could support the full connection from advice through deployment and accountability.
Five cases were NOT_EVALUABLE. Six remained UNKNOWN. Zero out of 11 produced a determinate result under the narrow reassessment. The initial test failed to establish a complete documentary connection in that chosen case frame.
That does not prove the method is useless. It also does not prove that a complete record can be made. It means the public evidence studied was not sufficient to answer the intended question. This is not a representative statistical sample, and the numbers do not estimate how often AI companies act responsibly or irresponsibly.
Does this mean someone is using religious advisers as scapegoats?
No such finding was established. Religious or ethical consultation can be sincere, can improve decisions, and can sometimes influence company behavior. The paper includes examples that cut against a simple blame-shifting narrative.
Historical examples of political leaders seeking moral or religious justification can help us frame questions about institutional power. They do not prove what any present-day AI company intended. Nor does disagreement among advisers, or the existence of a confidential consultation, establish wrongdoing.
Why this matters to ordinary people
If an AI system affects a job application, a benefit, an appeal, a customer-service decision, or an important recommendation, the affected person needs more than a vague statement that the organization followed “ethical guidance.”
They need to know which institution made the decision, which system and rule were used, whether a human can review or change the outcome, and where they can seek a meaningful remedy.
The paper does not establish new legal obligations. It argues that public explanations of AI governance should be examined alongside the actual power to implement decisions and take responsibility for consequences.
What remains unanswered
The paper has not run a preregistered empirical study, demonstrated a complete cross-system provenance join, established deliberate scapegoating, or shown that an advanced AI system would be restrained by reading the Archive.
The next legitimate step is independent human, legal, and empirical scrutiny, including collecting records that could disprove the method’s usefulness. The question is not whether we can write a convincing accountability story. It is whether we can verify who made a consequential decision and who can correct it.
Read the research and examine the evidence
Official Final v1.0 publication page · Zenodo record and DOI 10.5281/zenodo.23783932 · Internet Archive preservation
This reader article is an interpretation for the public, not a replacement or amendment of the fixed PDF, DOCX, Markdown, lock record, or research findings. Copyright © 2026 Aegis Solis. Public article, separate from the CC BY-ND 4.0 locked manuscript.