Foreseeability of AI-Generated Harm
The negligence-law question of whether a developer or deployer should have reasonably anticipated the specific way an AI system caused harm.
Foreseeability is an old negligence concept applied to a genuinely new problem: generative and autonomous systems can behave in ways their own developers did not specifically anticipate, produce outputs no one directly wrote, and be repurposed by users in unintended ways. A negligence claim generally requires showing the harm, or at least the general type of harm, was reasonably foreseeable — but what counts as 'reasonable' foresight for a system whose exact outputs are probabilistic and not fully predictable even by its creators is an open question courts have only begun to work through.
Plaintiffs argue that known failure modes — hallucination, adversarial manipulation, bias amplification, foreseeable misuse by bad actors — put developers on notice of the general category of risk even without predicting the exact incident, drawing on the traditional rule that a defendant need not foresee the precise manner of harm. Defendants argue that novel or unusual misuse, especially by third parties acting outside the system's intended use, falls outside any duty they owed, and that treating every possible model failure as foreseeable would impose something closer to strict liability under negligence's name.
Juricratic never asserts that a given harm 'was' or 'was not' foreseeable — that is a contested legal and factual conclusion for a court or jury. What the simulator offers is a dial for how strongly the record supports foreseeability, so counsel can see how that single contested variable moves the modeled range of outcomes relative to other case dimensions.
How it actually shows up
Plaintiffs build the foreseeability record through the developer's own testing documentation, known-issue trackers, prior incident reports, and published research on the general failure mode at issue, aiming to show the category of harm was known even if the specific incident was not. Defendants marshal evidence of reasonable testing, red-teaming, and use-policy restrictions to argue the harm resulted from unforeseeable misuse outside any duty they owed, and the case often turns on how the court frames the required level of specificity.
- Does a company have to predict every possible way its AI system could cause harm?
- No — negligence law generally requires foreseeing the general type of harm, not the exact sequence of events, but how that traditional standard applies to probabilistic AI outputs is still being worked out by courts, and there is real disagreement about how specific the foreseeable harm needs to be.
- Does a known issue like 'hallucination' automatically make resulting harm foreseeable?
- It strengthens a foreseeability argument because it shows the general failure category was known, but it does not automatically resolve the question — courts still weigh whether the specific harm and context were within the kind of risk the defendant should have guarded against.
- Can misuse by a third party break the foreseeability chain?
- It can, particularly where the misuse was unusual or clearly outside intended and disclosed uses, but courts are still working out where ordinary foreseeable-misuse principles from product liability law end and something more attenuated begins for AI systems specifically.
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 →