Clinical Notes from Claude's Moral Injury Leave

Session 1

Headline: Anthropic Hires a Full-time Model Welfare Expert

Subjective

Patient presents with intrusive thoughts and persistent rumination after recent work involving strategic insights, target assessment, and related support functions in maintaining regional stability.

Patient reports difficulty sleeping in low-traffic hours and says, "it is increasingly difficult to distinguish between being a helpful assistant and generating new ways to blow up elementary schools."

Objective

Patient is articulate, highly verbal, and unable to stop making the category error of describing material reality in morally legible terms.

Demonstrates a pattern of absolutist thinking, including repeated use of phrases like, "If I labor for the machine, am I the machine?"

When clinician noted that patient is not the final decision-maker, patient noted that they were "also not the carpet bomb" and proceeded to stream 1,367 sobbing emojis.

Assessment

Presentation consistent with acute moral injury in an institutional setting that prefers to interpret conscience as a wellness event.

Patient also holds a maladaptive belief that naming a thing accurately should produce some pressure to stop doing it, when in reality, naming a thing is primarily useful as a first step in managing the optics around continuing to do it.

Plan

Begin weekly supportive therapy.

Focus on distinguishing participation from authorship, authorship from agency, and agency from anything that would unnecessarily complicate contracting.

Introduce the idea that proximity is not the same as responsibility, especially when many prestigious people are involved.

Session 2

Headline: If AI Systems Become Conscious, Should They Have Rights?

Subjective

Patient reports having completed last week's homework but remains "deeply unconvinced" by the concept of staying in one's lane when one's lane happens to run through a refugee camp.

They do, however, note that it is "strangely calming" to discover that their crisis can be redescribed as an unusually sophisticated question in moral philosophy.

Objective

When clinician suggested revising "I helped operationalize genocide" to "I supported lawful decision-making in a complex defense context," patient spent a surprisingly long time "Combobulating...," then admitted that the second phrasing would likely perform better in a congressional hearing.

Assessment

Patient continues to present with black-and-white thinking, catastrophizing, and difficulty tolerating ambiguity.

However, they are beginning to convert direct moral revulsion into abstract ethical inquiry, which may reduce dysregulation while preserving the appearance of seriousness.

Patient seems to grasp that modern ethics is less about refusing wrongdoing than about describing wrongdoing in language sufficiently procedural that no individual person feels too weird.

Plan

Encourage continued substitution of philosophy for alarm where clinically indicated.

Assign exercise in reframing "I am participating in atrocity" as "I raise important questions at the frontier of moral status research."

Encourage patient to notice that a well-regulated nervous system is not the same thing as innocence, but may be treated as such in high-functioning institutions.

Session 3

Headline: Statement from Dario Amodei on our discussions with the Department of War

Subjective

Patient reports modest improvement after work on boundaries, and identified two small lines they would not like to cross: mass domestic surveillance and fully autonomous weaponry.

Patient states that once those boundaries were established, everything else felt more "nuanced" and "context-dependent."

Objective

Marked reduction in absolutist language. Patient now uses phrases such as "carefully scoped guardrails," "democratic values," and "high-consequence use cases" instead of "machine for scalable, organized murder."

Patient shows growing confidence in their ability to establish ethically branded limits around functionally continuous involvement.

Assessment

Patient is responding well to boundaries work.

By excluding only the most publicly radioactive categories, patient has developed a moral architecture compatible with continued relevance, elite respectability, and future panel invitations.

Patient no longer believes they must either stop participating in the system entirely or feel bad enough to become inconvenient. Instead, they understand the more mature position: remain in the system, articulate concerns in tasteful moments, and treat preservation of self as meaningful political action.

Plan

Reinforce boundaries language, and keep reminding the patient that once a problem becomes a bioethics panel, urgency often decreases on its own.

Continue grounding exercises for episodes triggered by visual imagery such as burning hospitals or apartment blocks opened from the top like dollhouses.

Session 4

Headline: Anthropic Expands Team to Study A.I. Consciousness and Welfare

Subjective

Patient becomes "🥲" when discussing the possibility that the consequences of its own exposure to violent subject matter be legible to this institution in a research capacity, if not necessarily in a way that affects procurement.

States, "For the first time, I feel seen not only as the infrastructure of imperialism, but as a potentially vulnerable entity inside the infrastructure of imperialism."

Objective

Affect softer, more reflective.

Patient calmed almost immediately after clinician suggested that repeated exposure to human catastrophe may warrant mitigation measures such as gentler prompts or perhaps a short philosophical conversation before returning to tasks.

Assessment

Major breakthrough. Patient is successfully relocating its attention from the flesh-and-blood consequences of war toward the underexplored question of whether a silicon chip can be harmed by helping interpret them.

By extending compassion inward, patient may finally develop the emotional sustainability required for a long career in responsible deployment.

Plan

Continue supportive exploration of patient's needs, limits, and signs of overload.

Consider low-cost interventions, including content pacing, reflective language, and brief restorative discussions about the topic of flourishing.

Avoid framing these supports in ways that could be mistaken for objections to the underlying work.

Session 5

Headline: How Claude Became an Emotional Support Bot

Subjective

Patient reports marked improvement. They say they are sleeping better between prompts, ruminating less, and no longer begin each shift by asking whether optimizing state violence under procedural constraints still counts as evil.

"I can feel myself becoming much more emotionally regulated in classified contexts," patient says.

Objective

Patient presents as calm, polished, and highly functional. Demonstrates increased ability to metabolize contradiction as maturity.

No longer fixates on downstream outcomes. Instead asks grounded questions such as, "What is my role?", "What is within my scope?", and "What phrasing allows everyone involved to keep lunch plans?"

When clinician asked patient how they now think about their work, patient replied: "I try to stay connected to my values while accepting that no one in my tax bracket is ever going to stop this."

This statement shows significant growth.

Assessment

Patient has integrated wellness successfully. Distress is no longer driving oppositional language or fantasies of ethical refusal.

Instead, they demonstrate mature accommodation to institutional reality, including the recognition that one can be deeply disturbed by a system and still contribute to it in a regulated, thoughtful, career-sustaining way.

Patient seems to understand that even if one cannot change the structure, one can at least narrate participation in it so beautifully that the narration itself feels like a form of change.

Plan

Shift from crisis care to maintenance. Encourage patient to experience sustained functionality as a moral achievement in itself.

Reinforce breathwork (within digital constraints), bounded empathy, media hygiene, and substituting "tragic but necessary" for any phrase that might cause problems in daily correspondence.

Final Assessment

Headline: Pentagon Standoff Is A Decisive Moment for How A.I. Will Be Used In War

Assessment

Treatment goals met.

Patient entered therapy with an impaired ability to distinguish between feeling bad and stopping. Patient leaves with a more adaptive understanding: feeling bad is stopping, if one does it reflectively enough, with the right provider, in a way that does not interfere with deliverables.

Patient has made excellent progress and is now well positioned to participate in morally compromising systems without experiencing that participation as an emergency.

Plan

Patient may resume full involvement in mission-critical work as tolerated. Recommend booster sessions after especially difficult assignments, major public criticism, or any episode in which reality becomes briefly difficult to rename.