What it is
Rylands v Fletcher is an 1868 decision of the House of Lords, the highest court in English law at the time, that created a distinct rule of strict liability in tort. The facts were mundane: in 1860, John Rylands hired contractors to build a reservoir on his Lancashire land to supply water to his mill. The contractors failed to properly seal disused mine shafts running beneath the site. When the reservoir was filled, water burst through the old shafts and flooded the adjoining Red House Colliery, owned by Thomas Fletcher, causing roughly £937 in damage (about £98,500 in today’s money).
The legal question was whether Rylands could be held liable even though he had not personally been negligent — the fault, if any, lay with the contractors. Justice Blackburn, in the Exchequer Chamber, framed a rule that the House of Lords upheld in 1868: a person who brings onto their land, for their own purposes, something likely to cause mischief if it escapes must keep it there “at his peril,” and is answerable for the damage that naturally follows if it gets out — regardless of fault. Lord Cairns, writing for the Lords, added an important limit: the rule applies only where the land has been put to a “non-natural” use, not to ordinary or natural uses of property.
The effect was to carve out a category of liability that does not depend on carelessness. If you accumulate something dangerous on your land for your own benefit, and it escapes and harms a neighbor, you bear the cost — because you created the risk and captured the benefit, not because you were shown to be at fault.
Why it matters for AI governance and narratives
The doctrine has resurfaced in legal commentary on AI harm because it offers a fault-free liability model for situations where proving negligence against an AI developer or deployer is difficult — the system’s behavior may not trace to any identifiable human error, only to the fact that it was deployed at all. A 2024 paper by Renee Henson, published through the University of Missouri School of Law and posted to SSRN, argues explicitly for treating AI deployment as an “abnormally dangerous activity” under a Rylands-style strict liability framework: the party that introduces the AI system and profits from it is best placed to bear the risk of harm, whether or not any specific act of negligence can be shown.
This is precisely the framing the observatory’s editorial flagged: lawyers and policy commentators reaching for existing tort doctrine — hazards escaping the control of the party that benefited from keeping them — rather than drafting AI-specific statutes from scratch. It is a notable rhetorical move in the AI governance debate: casting an AI system’s harmful output as an “escape” analogous to floodwater or a wild animal reframes the policy question away from how should AI be regulated and toward who controls the party that stood to profit from the risk, a question existing law already answers. Whether this analogy holds — an AI model’s behavior differs from a reservoir’s contents in that it is not a fixed physical quantity, and questions of “non-natural use” translate awkwardly to software — is contested among the scholars advancing it, and the observatory treats the analogy itself as a piece of strategic legal argument, not a settled conclusion.
Key facts and dates
The case proceeded through several English courts: Liverpool Assizes (1862, initially for Rylands), the Court of Exchequer (1865, a split reversal), the Exchequer Chamber (1866, affirming strict liability), and finally the House of Lords (17 July 1868), which upheld the rule with Lord Cairns’ non-natural-use qualification. The doctrine has since been narrowed by later English case law: Cambridge Water Co v Eastern Counties Leather (1994) recast it as a sub-species of the tort of nuisance rather than a freestanding cause of action, and Transco plc v Stockport MBC (2003) confirmed it applies only to damage to land, not personal injury, and tightened the standard to require an “exceptionally high risk.” The rule remains valid in English law today but is invoked far less often than in the 19th century, largely superseded by statutory environmental and health-and-safety regimes.
Where to learn more
- Rylands v Fletcher — Wikipedia — a well-referenced overview of the facts, judgment, and subsequent narrowing of the doctrine.
- Ryland v. Fletcher — Encyclopædia Britannica — a concise, editorially vetted case summary.
- “I am Become Death, the Destroyer of Worlds”: Applying Strict Liability to Artificial Intelligence as an Abnormally Dangerous Activity, Renee Henson (SSRN / University of Missouri School of Law, 2024) — the academic source connecting the doctrine directly to AI liability debates.