Who owns a generated work is unsettled, and the reasons are older than the technology

The framework assumes a person made decisions

Systems of authorship rights were constructed around an idea of creative choice: somebody decided what to include, how to arrange it, what to leave out, and that decision-making is what the protection attaches to. Tools have always been involved — a camera, a synthesiser, a word processor — and the tool was never the author because it made no choices.

Generated output strains this because the choices are distributed strangely. A person decided what to ask for and which result to keep. A system produced the specific arrangement of pixels or words. A very large number of other people made the material the system learned from. None of those roles matches the one the framework was built around.

This is not a new kind of difficulty. Similar arguments arose over photographs, over works produced by processes with a random element, and over compilations. Each was resolved eventually, usually by finding the human decisions and asking whether they were substantial enough. The current case is harder because the decisions are thinner.

Jurisdictions have not converged, and that matters practically

Some legal systems require a human author for protection to exist at all, which points towards purely generated output being unprotected. Others have long-standing provisions for works produced by a computer, drafted decades ago for entirely different circumstances, that may or may not fit. Others have not addressed the question directly.

Guidance issued by registration authorities in several countries has tended towards a middle position: material generated without meaningful human control is not protected, while a work incorporating generated elements arranged and selected by a person may be protected as to those human contributions. Where the line falls is exactly what is disputed.

Because the answers differ by country and are actively developing, none of this constitutes advice about any real situation. Anybody with a commercial decision resting on it needs a professional opinion in the relevant jurisdiction, and should expect that opinion to carry caveats.

Unprotected is not the same as safe to use

A frequent misreading treats the absence of protection as a clean outcome. It is not. If output cannot be protected, then nobody can prevent anybody else from using it, which is a problem for anyone whose business depends on exclusivity. A brand identity that cannot be defended is worth considerably less than one that can.

Separately, the question of whether output infringes something else is entirely distinct from whether it is protected. A generated image closely resembling an existing work raises the same questions any close resemblance raises, regardless of how it was produced. The two issues get conflated constantly and they have no logical connection.

Providers of these systems have responded with contractual assurances of various kinds, which are commercial arrangements between two parties and do not change what the law says. They shift who bears a risk rather than removing it.

The argument about training is a different argument

Running alongside the authorship question is a much larger dispute about whether training on gathered material required permission. Legal systems differ on whether analysis of a work is the kind of use that needs authorisation, and several have exceptions for text and data analysis with conditions attached.

These two arguments are frequently merged in public discussion and they are separable. It is coherent to hold that training was permissible and output is unprotected, or that training required licensing and output belongs to the person who prompted it. The combinations are all defended by somebody.

What connects them practically is that both are being decided now, by courts and legislatures working from analogies to earlier technologies. Analogical reasoning is how legal systems handle novelty, and it produces answers that depend heavily on which analogy is accepted.

Where this leaves the people affected

For working artists and writers the immediate issue has rarely been ownership of generated output. It is that a body of work can be imitated closely enough to substitute for commissioning the person who made it, without anything identifiable being copied. Style has never been protected, for good reasons, and the cost of that principle has changed.

Proposals for a remedy include rights over a recognisable personal style, compulsory licensing schemes, and disclosure requirements about training material. Each has serious objections — style protection in particular would restrict human artists learning from each other, which is how the field has always worked.

The honest summary is that a framework built for one situation is being applied to another, that the results so far are inconsistent, and that anybody claiming to know the settled answer is describing their preferred outcome.