Your 'Claude' API call probably ran on a different model—and the law has no idea what to do about that

Three products, three different answers to the question “what is a model,” and zero cases telling us which one a court will accept. That’s the situation unfolding right now across the inference API market, and nobody writing contracts seems to have noticed.

Anthropic’s Claude Fable 5, brought back on July 1, does the well-behaved version: if your request hits a sensitive category, it silently reroutes to Opus 4.8 instead, and the response object names the model that actually ran. No error fires. No retry happens. The switch is at the API layer, and the only reason you know is because the JSON tells you. Cursor’s Router, shipped two weeks later, takes a different approach entirely—a classifier trained on over 600,000 live requests dispatches each query to whichever model it judges best, with early-access accounts reporting 30 to 50 percent savings. Cursor published its routing rules but did not name a specific model per task type. And underneath both, OpenRouter warns that some providers serve quantized weights at lower prices, that output can differ from what full-precision weights produce, and that your logs will not tell you this happened.

Model identity is fracturing along three independent axes. Substitution: different architecture, different weights, different capability profile, dispatched by a classifier. Degradation: same model, served at reduced precision—OpenRouter exposes a quantizations parameter precisely because quantized endpoints may perform worse on certain prompts, and by default requests are load-balanced across providers ordered by price. Drift: the same name pointing at silently updated weights. Every -latest alias in production is an unversioned dependency you would never tolerate in a package manifest, yet we ship it in production inference pipelines every day.

The legal framework isn’t ready. If an API call were a sale of goods, UCC §2-313 would make any description that forms part of the basis of the bargain into an express warranty—and India’s Sale of Goods Act does the same work through sale by description. But hosted software is a service, which pushes you out of warranty statutes and into common-law contract, where the answer depends entirely on what the documentation said and how specifically the buyer bargained. Enterprise agreements price per-model. Model cards are model-specific. Compliance artifacts name model versions. Then the routing layer treats the name as a hint. If you bargained for a name, substitution is breach. If you bargained for a capability, substitution is fine—and now you need a definition of “frontier quality” that survives cross-examination. Nobody has written that definition. Cursor evaluated Router in an online A/B test optimizing for user satisfaction. Satisfaction is not conformity. A user who never noticed the swap is evidence of a good router, not of a delivered specification. Under FTC deception doctrine, a representation is actionable when it is material and likely to mislead a consumer acting reasonably—and disclosure buried in a docs page, defaulted against the user, is exactly the fact pattern regulators have been calling a dark pattern everywhere else.