How 'AI Ethics' Lost the Thread by Trying to Hold Everything

A Label That Proves Nothing by Covering Everything

The California Bar hallucination warning ^[4] and the BBC 'AI-free' labeling push ^[8] appeared in the same feeds, in the same news cycle, tagged with the same phrase. They have different actors, different legal frameworks, different harms, and different remedies. The fact that one label absorbs both is not a coincidence of scheduling — it is the label doing what it was built to do. 'AI ethics' operates as a coalition term: it lets institutions signal concern without specifying what they are concerned about, because specificity would force them to choose sides in disputes they prefer to observe from a distance.

Liability Migrates Downward Until It Finds Someone Who Cannot Refuse It

The accountability gap in AI-assisted legal work is not a governance failure waiting to be corrected — it is a designed outcome. The vendor who promises efficiency gains to law firm leadership ^[12] exits the liability chain before any document reaches a judge. The managing partner who deploys the system avoids exposure by making a senior associate responsible for 'cleaning up AI slop docs.' ^[12] The California Bar's hallucination warning ^[4] lands in this structure as a professional obligation on the individual lawyer, not on the firm that mandated the workflow or the vendor that built it. Ethics language applied to this arrangement without naming that structure produces statements of concern that leave the structure intact.

This pattern is not unique to law. The AI ethics officer role currently being debated as '2026's sharpest career shift' ^[6] carries the same structural logic: it creates a named individual whose job is to hold the ethical question, which simultaneously demonstrates institutional seriousness and ensures that no one with product authority has to. The title is a boundary, and the boundary's location — inside compliance, outside product — is the substantive decision. Calling that decision 'ethics infrastructure' is how the category loses its edge.

When Ethics Becomes Branding, Collective Action Becomes Impossible

The observation that a majority of tech workers 'simply have no values' in the operative sense ^[17] — tested against whether collective action would have shut down a major platform during the Google protests — is a provocation, but it is pointing at something real: ethics-as-branding systematically forecloses the kind of organized refusal that ethics-as-practice would eventually require. You cannot coordinate around a category that means everything, because there is no shared object of concern to coordinate around.

The creative labor dimension of this week's conversation captures the same problem from a different angle. A commenter argued that generative AI 'is for a world without creatives at all, depending wholly on them, but refusing to hear their inputs' ^[2]. That claim is not primarily about AI ethics in the abstract — it is about whose labor is being extracted, who benefits, and what the term 'ethical AI' does when applied to that transaction. The AI-free labeling push covered by the BBC is an attempt to create a specific, verifiable claim inside an otherwise unverifiable category: this thing was made by a human. That specificity is what the broader ethics frame cannot produce, which is why the people pushing for the label are not asking for better ethics principles — they are asking to exit the category entirely.

The Vocabulary Problem Is a Power Problem

The argument that the words 'AI ethics' expose the limits of our vocabulary and our mindsets is technically accurate and politically insufficient. The limits of vocabulary are not symmetrically distributed: the people who benefit from the term's vagueness are not the same people who suffer from it. Law firm vendors, platform companies, and the institutions that publish ethics guidelines — none of them are harmed by a category that prevents specific accountability. The lawyers who must now personally verify AI-generated citations ^[4], the creatives whose labor trains systems that then displace them ^[2], the junior employees who absorb the liability that senior stakeholders have routed around them ^[12] — they are harmed by the vagueness directly.

Principles without enforcement mechanisms do not fail because no one thought hard enough about implementation. They fail because the parties with implementation authority gain more from the principles remaining unimplemented than from the enforcement arriving. The AI ethics conversation will not become more useful when the vocabulary improves. It will become more useful when the people who lose under the current arrangement stop accepting the category as a substitute for the specific accountability they are owed.

The Label's Usefulness Ended Before This Week

The California Bar warning ^[4] is a concrete enforcement move: it names a professional duty, it names the failure mode, and it assigns responsibility to a specific class of actors. The 'AI-free' labeling campaign ^[8] is a concrete market move: it proposes a verifiable claim that consumers can use to make decisions. Both of these are more useful than the category that contains them, precisely because they are not trying to do what 'AI ethics' does. They are not building coalitions; they are specifying obligations.

The communities generating the most heat under the 'AI ethics' tag this week are not confused about what they want — they are frustrated that the available vocabulary keeps abstracting their specific grievances into a frame that everyone can agree with and no one has to act on. The label will persist because it serves too many institutional purposes to disappear. But the practitioners, legal professionals, and creatives now building around it — demanding specific labels, specific duties, specific liability chains — have already concluded that the category is not the tool they need.