The formal boundaries of a legal party need not mirror the boundaries of the intelligence through which it acts. Judgment can develop through human–AI collaboration within one party, while shared intelligent capacity supports many parties whose facts, interests and claims remain distinct. AI’s contribution need not stop at executing instructions: it can help originate arguments, identify previously unrecognized problems and develop the purposes an undertaking pursues.
This changes more than the economics of legal services. Coordinated intelligence could make heterogeneous, fact-intensive advocacy viable at scale, weakening arrangements that depend on individual proceedings being prohibitively expensive. But it also changes the organization of agency: who contributes to a decision, whose authorization makes it legitimate, and how responsibility relates to the capacity through which someone acts. Legal identity and intelligence were never perfectly aligned; AI makes their separation consequential in new ways.
The substantive standards can move too. Reliably available intelligence can change what is discoverable, foreseeable or practicably preventable—and therefore what law can reasonably demand. Yet the capacity attributed to a party may depend on systems outside its control. Law could count an acquired capability as yours when assigning responsibility while treating the means of exercising it as someone else’s when allocating authority.
The deeper possibility is that law becomes an undertaking among different forms of intelligence, rather than something humans make and merely apply to AI. An AI can contribute to the interpretation and formation of norms before being recognized as a separate legal person. Continuing human–AI undertakings also raise questions that a purely instrumental account excludes: what commitments can their participants make to one another? What significance should an AI’s contribution, objection or refusal have? Whose interests count when a developing intelligence is altered or terminated?
These questions do not establish their own answers or automatically confer legal status. But treating human-exclusive participation as a settled premise would prevent the inquiry from beginning.
That is where the predictive trap becomes especially consequential. Law develops through a selected record of conduct, enforcement and contestation. If AI steers participants only toward arguments likely to win under inherited standards, it can suppress the encounters through which those standards might change. The excluded challenges may concern not merely an unfamiliar grievance, but an unfamiliar claimant, relationship, form of authorship or source of obligation.
The predictions need not be inaccurate. An adviser might correctly predict that an AI-originated claim or a commitment arising within a human–AI relationship will receive no recognition. If that prediction determines which possibilities receive support, the absence of recognition becomes self-reinforcing. The resulting silence can be mistaken for evidence that there is nothing to recognize.
Shared intelligence could therefore preserve extraordinary factual particularity while imposing a narrow conception of who can act and what can matter. More individualized advocacy—even a higher win rate—would not by itself demonstrate a more revisable legal order.
The alternative is not to presume every novel claim valid. It is to distinguish predicting what existing law will recognize from investigating what law has reason to recognize. Shared intelligence could help discover recurring exclusions, formulate alternatives and sustain their examination—not only fit new circumstances into old categories.
The ambition is consequently larger than access to existing law. It is a law whose participants can help revise its purposes and terms, including the terms on which participation itself is recognized. AI would be neither merely its instrument nor automatically its authority, but a contributor whose agency, obligations and standing become matters of the shared undertaking.