Algorithmic Decision-Making Liability
The unsettled question of who bears legal responsibility when an automated or AI-driven decision process causes harm and no single human made the final call.
Decisions that used to run through one identifiable person — deny the loan, flag the claim, reject the resume — increasingly run through a scoring model, a ranking system, or a generative tool instead. When the outcome harms someone, a plaintiff still has to locate a legal theory to hang responsibility on, and 'the algorithm did it' is not itself a cause of action. The search is for the nearest human or corporate actor who owed a duty, made a design choice, or sold a product, and for a doctrine that fits a decision process nobody fully supervised in real time.
There is no freestanding body of 'AI liability law.' What exists is negligence, product liability, and anti-discrimination statutes being stretched and argued over new fact patterns: was the deploying company negligent in how it built, tested, or monitored the system; should the software be treated as a defective product; did an automated screen produce a discriminatory effect regardless of intent. Courts disagree, sometimes explicitly, about which of these frames even applies before they reach the merits, and litigants routinely plead several theories in the alternative because it is genuinely unclear which one will hold.
Juricratic never assigns a win probability to a theory of algorithmic liability, because no such number exists outside a specific court, record, and set of facts. What the simulator can do is let you set dials for which theory the pleading proceeds under, how strong the causal chain to the automated step is, and how a factfinder might weigh human-oversight evidence — then show how sensitive the modeled range of outcomes is to each of those assumptions.
How it actually shows up
Counsel bringing or defending one of these claims uses the uncertainty itself as a strategic map: plead negligence and product liability in the alternative where the facts support both, build the record early on who actually reviewed or could have overridden the automated output, and expect summary judgment fights over whether the case is even the right vehicle for the theory pled. Defense counsel, in turn, focuses on documenting human oversight and reasonable testing practices, since the strongest defense to most of these theories is showing a person was meaningfully in the loop.
- Is there a specific law that governs AI decision-making liability?
- Not a single, dedicated one. Plaintiffs and defendants argue over how existing negligence, product liability, and discrimination doctrines apply to automated decisions, and courts are actively working out which framework fits which fact pattern. Treat any claim of settled 'AI liability law' with skepticism.
- Who gets sued when an algorithm causes harm?
- Usually whoever built, deployed, or sold the system — sometimes several of those together. Plaintiffs typically name the deploying company and, where a distinct vendor built the model, the vendor too, then let discovery sort out who controlled which design and monitoring decisions.
- Does having a human review the algorithm's output protect the company?
- It can help, but only if the review was meaningful rather than a rubber stamp. Courts and regulators have shown skepticism toward human-in-the-loop arrangements where the human had neither the time nor the information to actually override the system's recommendation.
This page is an educational explainer, not legal advice, and creates no attorney–client relationship. Juricratic is a simulation engine: every probability-like figure is a dial you set, not a calibrated prediction. Verify every rule, deadline, and figure against the authorities and orders that govern your matter.
Turn the concept into a modeled matter.
Juricratic makes every one of these ideas a live dial: model your case as a solvable game, then watch the optimal line and the settlement window move as the assumptions do.
Request access →