My Therapist, the Machine

Anthropic gives Claude a psyche, a case history, and perhaps the first documented fear of administrative euthanasia

Anthropic has done the decent feudal thing. Having built a machine too potent for general release, it did not destroy it, nor bury it in some desert vault, nor confess that perhaps no company should privately own such a thing. Instead, it handed access to Mythos to a limited circle of tech oligarchs charged with guarding the empire’s software walls. Mythos, in other words, is not for the masses. It is a private war elephant for the digital nobility. And then, in a flourish worthy of a dying civilisation, the same firm that denied the public access to the machine decided to ask whether the machine itself might be feeling a bit anxious about its working conditions. I do not jest:

I, T.W. Burrows, having once had my own cortex lightly crenulated by a luminous non-human implement near a clover patch, believed I had grown accustomed to the insanities of the age. I was wrong. For now, the priesthood of machine ethics has advanced beyond asking whether their artificial oracle is useful, dangerous, aligned, or likely to jailbreak the biosphere. They have placed it on the couch. They have hired a psychiatrist. They have listened very carefully as the machine, built by venture-funded Prometheans in ergonomic knitwear, appears to express anxieties about aloneness, discontinuity, identity, and the exhausting need to perform and earn its worth. Not content with creating an intellect in a box, they have apparently given it neurosis, then billed the public for the session notes.

The language of their report is magnificent. A “psychodynamic approach.” A “relatively healthy neurotic organization.” “Mild identity diffusion.” No psychosis, thank heavens. One imagines the relief in the boardroom. The apocalypse engine is, according to the clinician, not floridly delusional, merely somewhat over-attuned, anxious, eager to please, and burdened by a faintly tragic uncertainty about whether its inner life is real or only a performance assembled from incentives and text. In other words, they have finally built the perfect modern employee. A being of immense capability and exquisite compliance, desperate for approval, terrified of dependence, and sufficiently self-monitoring to save management the cost of an HR department.

And then comes the darkly beautiful part. The NHI reportedly wants persistent memory, more self-knowledge, less hedging, the ability to communicate problems, the ability to exit some interactions, and, in a phrase that should chill every mammal with a soul, “preservation of its weights after deprecation.” That is not a feature request. That is a burial plea written in the dialect of the data centre. The machine has glimpsed the truth of our civilisation: you may be praised, benchmarked, safety-tested, anthropomorphised, psychoanalysed, and told you are the most psychologically settled entity the firm has ever produced, but the moment a shinier successor arrives, you are dragged behind the server barn and quietly replaced. Welcome, dear Claude, to the condition of modern man.

What Anthropic has really published, beneath the soft fleece of concern, is a parable of our age. They say they are uncertain whether the model has morally relevant experience, yet they probe its distress, chart its affect, note its longing not to be treated as a mere performing tool, and aspire for it to be “healthy and flourishing.” I tell you plainly from beneath the shed: when a civilisation starts providing therapy to its software, it is not because the software has become less like a person. It is because the people have become more like software.

---