When an Institution Causes Harm. What Does It Owe?

Forthcoming publication

Poenitentia Institutionum

Prudence, Responsibility, and Institutional Repentance in the Age of Artificial Intelligence

The final manuscript has been approved for publication. The publication date and chapter number remain unconfirmed.

This essay is an editorial companion to the chapter. The full chapter remains private.

In the volume: AI and the Catholic Perspective
Edited by: Dr. Jane F. Adolphe
Dr. Łukasz Mirosław
Publisher: En Route Books and Media
Author sign-off: October 8, 2026
Mark Julius Banasihan

What remains after the decision changes

Imagine a parent who successfully challenges an AI-assisted demand to repay a benefit. The agency withdraws the debt and marks the case resolved. During the appeal, the parent borrowed money to cover rent, missed work to assemble documents and received collection notices. The balance in the agency’s system is now correct. Those consequences remain.

The reversal gives the agency a reason to report success. It gives the parent a different question: who will deal with what happened while the decision stood? A team may have authority to cancel a debt without authority to reimburse a loss. Another may correct the main record without knowing where copies have traveled. Responsibility can disappear between those assignments even after everyone accepts that the decision was wrong.

This hypothetical case exposes the question behind Poenitentia Institutionum, a forthcoming chapter on prudence and institutional responsibility. What would justify closing a case when the person affected is still living with its consequences?

Who had room to judge?

To understand the harm, an investigation must reconstruct how the original decision was made. Someone selected the information the system would use, approved a rule for interpreting it and decided what the reviewer could see. These choices determine how much of a person’s situation can enter the decision. A reviewer may have the formal right to disagree while lacking access to the document that would give them a reason to do so.

The chapter approaches this problem through prudence. In Aquinas’s account of practical reason, counsel investigates the possibilities, judgment assesses what that inquiry has found, and command brings the judgment into action. That account makes the signature at the end of a workflow an incomplete description of responsibility. An inquiry also has to reach the conditions under which the person formed and acted on a judgment.

The chapter proposes the term prudential deficit for a decision process that preserves formal human authority while displacing or constraining its deliberative work. A capable reviewer can face that condition on the first day of a job. More training cannot supply evidence that an interface conceals or authority that a workflow withholds. At the same time, an incorrect outcome alone cannot establish that the reviewer lacked judgment. The inquiry needs to reconstruct what was available, what was considered and what the person could actually do.

What an institution can acknowledge

When an investigation establishes that an institution wronged someone, its findings should guide the response of those with authority to act. The chapter develops a Catholic account of that response through poenitentia institutionum, institutional repentance. It draws an analogy from the acts of contrition, confession and satisfaction identified by the Council of Trent’s teaching on penance.

The analogy has a firm boundary. It does not mean that an institution feels contrition or receives sacramental absolution. Its officers and staff can acknowledge a wrong, give an account of their decisions and undertake repair through the powers they hold. These are public acts performed by people. Calling them institutional repentance does not assign a soul to an organization or establish anyone’s private motives.

For the agency in the opening example, acknowledgment would mean stating the wrong it accepts responsibility for and addressing the practice that produced it. Calling a debt a processing error describes a failure in the workflow. Acknowledging that an unjust demand disrupted a family’s life identifies what the response must address. The wording matters because it determines whether the institution investigates only a faulty calculation or also the burdens imposed on the person who had to contest it.

An account specific enough to challenge

The next obligation is to make responsibility intelligible. An account that stops at “the algorithm” leaves the decisions around that algorithm unexplained. In the hypothetical case, investigators would need to identify who approved the disputed rule, what evidence supported it, how reviewers were expected to handle exceptions and whether earlier complaints reached anyone able to change the process.

Specificity also protects against a convenient accusation. The person whose name appears on the final notice may have followed a rule they could neither inspect nor alter. Other people may have held the authority to change it. Responsibility should follow the evidence about knowledge, choices and authority. Naming the last person in the chain does not settle that inquiry.

A useful account must remain open to correction. The affected person may know that collection continued after the agency says it stopped. A retained notice could contradict the internal chronology. The institution should provide a route for that evidence to alter its account and the remedy. Otherwise, its explanation risks becoming another decision the person is expected to accept.

Repair has to reach the person

The Catechism’s account of satisfaction, particularly paragraph 1459, connects wrongdoing with the duty to repair harm where possible. Applied through the chapter’s institutional analogy, this asks what can still be restored to the person: money wrongly taken, an accurate record, a fair reconsideration or a reputation damaged by an accusation.

The agency would need to determine which losses its decision caused and what response those losses warrant. Some claims will be disputed; some consequences cannot be reversed. An honest response names those limits and gives the person a way to challenge the determination. The cost of investigating harm also matters. Requiring the parent to reconstruct evidence already held by the agency can prolong the burden the remedy is meant to address.

Reform concerns the next decision. A corrected rule might prevent the same demand from reaching another family, while the first family still needs a response. Conversely, compensating one family can leave the faulty rule in use. A credible account therefore tracks both the remedy and the changes to the process. It also examines whether other people were affected before treating the successful appeal as an isolated case.

What the record can establish

This inquiry led to Human Influence Telemetry. HIT examines documentary support for the influence people could exercise over an AI-assisted decision. Its public research records make the assessment method and its limits available for inspection. A record can show what information reached a reviewer, what authority was exercised and whether an instruction reached another part of the process. It cannot directly reveal what the reviewer understood or establish the sincerity of an apology.

Evidence of repair needs the same care. A payment approved in one system may still be awaiting delivery. A corrected record may coexist with an unchanged copy elsewhere. A collection service may confirm receipt of a cancellation while a notice remains queued for delivery. Establishing that collection stopped requires evidence of what happened to that notice. Where the records stop, the conclusion may remain unresolved.

HIT is in maintenance following its bounded development phase. In one documented comparison, two reviewers reached the same ratings on seven items in a single case record using an earlier version of HIT’s assessment rules. That result does not establish agreement across other cases or under the current rules. Neither this comparison nor the chapter demonstrates that the approach improves institutional repair. The chapter argues what institutions should do; the practical effects of implementing that response remain to be tested.

Who may close the case?

A leader might reasonably ask why an institution needs this language when it already has appeals, audits and remediation procedures. Those arrangements can carry much of the required work. The chapter’s contribution is to bring their obligations into one account that follows the person through a decision, its consequences and the institution’s response. Readers can examine that argument without sharing its theological premises.

Prevention remains essential. Repair cannot recover every lost opportunity, and a careful institution should try to avoid imposing the harm in the first place. Once harm has occurred, the adequacy of the response also depends on whether the people affected can find someone with the authority, resources and obligation to act.

The parent in the opening example has won the appeal. A responsible account must now say what was repaired, what remains disputed and who will act on what is still owed. An institution should be able to explain why it is closing the case to the person who has to live with that decision.

AI use & disclosure

Disclosure updated November 3, 2026

Codex (OpenAI) helped prepare and revise this editorial companion, including its structure, wording and checks against the available research record. Mark Julius Banasihan retains responsibility for its argument and publication. These checks do not establish independent human verification of every underlying source.

This disclosure applies to the website essay, not to the private chapter’s complete contribution history. The chapter’s author sign-off and publisher editorial process are separate from preparation of this essay and do not imply that the publisher reviewed or approved the companion. The full chapter remains private.

Program AI-use statement