Anthropic and the Vatican show what breaks when AI is described as a being

The Register's May 27, 2026 article is not a conventional report about a new model, API or funding round. The supplied signal note already frames it as an opinionated, satirical piece with no tangible technical breakthrough. That does not make it useless. It points at a recurring weakness in the AI debate: both the industry and its critics often talk as if they are discussing beings, not systems.

The available context centers on a Register article about an Anthropic co-founder, a papal speech about AI and the old metaphor of a “ghost in the machine.” The limits matter. We do not have a quoted speech, a technical paper or a new Anthropic research release in the supplied material. We have a provocation and a blunt framing line: AI is unnatural, not intelligent and not human. That is enough for an editorial analysis, but not enough to pretend this is a major industry event.

The problem starts when the word “intelligence” is treated too literally. Systems developed by companies such as Anthropic can produce fluent language, classify content, assist with code and handle complex workflows. None of that proves intent, consciousness, responsibility or subjective experience. Once that distinction disappears, the public conversation becomes a theater of metaphors: the model “thinks,” “wants,” “knows,” “fears” or “understands.” Those words can work as casual shorthand. They are much worse as a foundation for policy, safety and accountability.

The Register's original piece appears to lean into that discomfort. A papal speech about AI brings a moral and anthropological register; Anthropic brings the industry register of safety, scale, model control and public trust. When those registers collide, the result can be large language and small precision.

That is why two questions need to stay separate. The first is legitimate: how should society govern systems that already affect work, education, information flows and security? The second is a bad leap: the idea that those questions must be debated as if models were new kinds of persons. For the first question, there are concrete tools, from the NIST AI Risk Management Framework to debates about accountability, transparency and testing. For the second, there is usually rhetorical fog.

The useful TECH&SPACE reading is not to mock religious language or industry figures who reach for grand analogies. The useful point is descriptive discipline. If a model hallucinates, say it generated an inaccurate or unsupported output. If a model answers well, say it produced a useful response within a given context because of its architecture, training and deployment conditions. If a system creates risk, name the risk: privacy, security, bias, automated decision-making, misinformation or organizational dependence on a vendor.

That is less spectacular than a ghost-in-the-machine story, but it is more useful. The AI debate is not short on mysticism. It needs colder inventory: what the system is, what data and constraints shape it, who is accountable, where the tests are, where independent checks happen and what follows when the system fails. Without that, every claim about “intelligence” becomes a mirror for human projection rather than an analysis of technology.