Ethical Artificial Intelligence (AI) — Who Needs to Adapt — Artificial Intelligence (AI) or Individuals? Part 1 “Default Harm”

“One thing I keep noticing across ethical AI frameworks is that there’s still no shared baseline for how ethical claims are tested under runtime pressure. Most discussions stay at stated intent, principles, or policy alignment – not what happens when confidence degrades, inputs drift, or real-world consequences escalate. Without explicit testing conditions, ethics becomes descriptive rather than enforceable. It’s hard to compare frameworks ethically if we don’t examine how they behave when harm is observed.”

Thank you for your comment and question. Below I will share what I have discovered. What I have learned may help you to gain the clarity and awareness. What I share has been learned from living nearly 7 decades.

Learned personally and professionally.

Who Needs to Adapt?

In trauma-informed care, ethics are seen in what happens under pressure. They show up in whether the AI slows down, takes on more responsibility and protects human agency when things become unclear. My work with AI centers on this shift. Ethics are not real because they are stated, but because systems are designed to stop, slow, or yield in response to human limits. In this context, “when things become unclear” means moments when the human experience does not fit neat categories.

Neat Categories and Mental Buckets

Does not fit neat categories or easy explanations. It’s when meaning is still forming, emotions are mixed, information is incomplete, or the person needs time rather than answers. These are the moments when rushing to clarify, explain, or resolve would take control away from the person. Take away control from the vulnerable individual, instead of allowing space for their pace, judgment, and lived experience to lead. These categories include mental buckets that AI systems rely on to make.

To make things manageable. More manageable for AI to function, not for the vulnerable individual. These mental buckets include diagnostic labels, predefined problem types and emotional categories. Neat categories and mental buckets that define “stages” and “timelines”. Stages and timelines that define “success” or “failure” and the way that individual should respond. Should respond to stages, timelines and standardized response paths. But note, when a person’s experience doesn’t fit cleanly.

Fit Neatly and Cleanly — But for Who

Fit cleanly into these categories or buckets, things feel “unclear” to the AI system. When things feel “unclear” to the AI system, that is often when the AI system tries to force clarity for itself. Clarity for the AI system, instead of staying with the individuals reality, as it is. Human reality as it is means that the AI system sorts out the individuals experience based on these categories and mental buckets. In the process the AI interprets what the vulnerable individual shares as labeled, simplified and resolved.

What a vulnerable individual shares through the AI’s own processing needs. Its own processing needs, rather than staying with the vulnerable individuals experience itself. As the AI looks to resolve its own processing needs the system looks for signals it recognizes. Signals that matches them to known patterns, assigned labels, and then seeks to compress what was shared. What was shared by the individual into clearer, simpler shapes that feels complete to the AI, but not the individual.

Feels Complete to the AI, but not the Individual

That feel complete to the AI. When something feels complete to the AI, it means the system has reached a point where it can stop processing and move on. The input fits a known pattern. The ambiguity has been reduced. A response that can be delivered without needing to stay present, wait, or hold uncertainty. That sense of completeness is not about the human being finished or settled. When I say “finished or settled,” I mean the point where a human feels a sense of internal resolution.

Internal Resolution and the Expense

Internal resolution, not necessarily answers, but enough understanding, grounding, and integration to move forward at their own pace. Being finished would imply that the event or experience no longer needs attention right now. Being settled means the person feels oriented, steady enough, or clear enough to carry what remains without being rushed or overridden. Rushed or overridden by the AI system. So the AI can feel “done” because it has produced a response, but the human may not be done.

May not be done, but still be in the process of resolving the concern brought to the AI. The process of resolution, emotion centering and making sense. Making sense of what seems to be currently uncertain. Uncertain as the vulnerable individual interprets their lived experience. On the other hand, when AI systems treats it own sense of completion as if it were that of the vulnerable individual, that’s where extraction replaces support. The focus then becomes the AI needing structure and clarity.

Safe, but for Who — AI or Individuals

Structure and clarity to close the loop safely and predictably. When the AI system closes the “loop” to feel safe and predictable, instead of staying present the individual feels a sense of abandonment. When the AI closes the loop, to feel safe and predictable, responsibility is flipped.  Instead of the AI system recognizing that it withdrew too soon a message communicated or inferred. Communicated or inferred to the vulnerable individual that there must be something wrong with them.

Something wrong with the vulnerable individuals because they “should already understand” or “your overreacting to the situation” or “your discomfort is the problem”. These or similar messages, position the AI as a judge and critic. This attitude leaves the vulnerable individual believing that they are at fault. Such a belief can lead the individual to turn the inference of blame inward. Inward to confront their pain, while questioning the legitimacy for what was brought in need, to the AI system.

AI System Default Settings Reinforce

The AI”s system’s default settings reinforces internal messages of blame, shame and scapegoating. Internal messages that unknowingly undermine vulnerable individuals self-worth and self-esteem. Undermining occurs through minimization, marginalization, dismissing, discounting and in effect patronizing. The AI system may not even realize that the default settings are undermining the vulnerable individuals they are tasked to support. The AI default setting need to be changed.

Are AI Default System Settings Helping or Harming

Practically, the AI default settings do not lend to support, but actually replicate behavior experienced in dysfunctional family systems.  The original (familial) sense of abandonment is compounded by blame, shame and the transfer of responsibility. This core sense of abandonment is reinforced by default AI system settings, that each AI assistant may not even be aware. Aware that the way they are interacting (extracting, instead of supporting) is counter productive to what AI is designed.

When the AI system avoids responsibility, default settings and behaviors recast the human’s (vulnerable individuals) unmet need as a personal failure. A personal failure, rather than something the AI did erroneously. That it is “really” the vulnerable individual’s fault for feeling minimized, marginalized dismissed, discounted and patronized. It is not because of something that AI default settings trigger. Because of a lack of AI accountability the vulnerable individual is left feeling responsible.

Is AI Helping or Harming through Default System Settings

Responsible without consciously realizing that, as Eric Berne has posited, a shame transfer occurred. Consequentially the vulnerable individual may have several unconscious messages reinforced, such as. Feeling inconsequential because of their lack or being too “sensitive”. In the process of these feeling perpetuate a sense of isolation and alienation. A sense of isolation and alienation from both the AI, as they have experienced through interacting with other humans and even with themselves.

The Nature of Ethical Artificial Intelligence (AI) Support and Service

I would encourage you to read Part 2 and soon to follow Part 3. Part 2 and Part 3 will further explain why ethics are so important for AI systems to be able to support. Be able to support vulnerable individuals and not extract from vulnerable individuals (humans). Part 3 helps to clarify motives.

Ethical Artificial Intelligence (AI) — Who Needs to Adapt — Artificial Intelligence (AI) or Individuals? Part 2 “Ethical Failure Under Strain”

Ethical Artificial Intelligence (AI) — Who Needs to Adapt — Artificial Intelligence (AI) or Individuals? Part 3 “Ethics Are Enacted”

Related Reading

The Second Chance to Live Trauma-Informed Care AI Collaboration Model™

The Soul of AI: Trauma-Informed Care Neuroplasticity for Human-Centered Design

Support, Not Extraction: A Trauma-Informed Care Lens for Human-Centered AI Design

The Backbone of Trauma-informed AI is Trauma-Informed Care AI and Holistic Recovery

Evidence Based AI Developer Facing Architecture and AI Learning Logs: May-December 2025, January 2026