Law / Product and AI liability
Product and AI liability law, instrument by instrument
Law on who pays when a product, software or an AI system causes harm: no-fault liability for defective products, software included in the European Union from December 2026, rules written for harm caused by AI, the presumptions and disclosure orders that help a claimant prove a defect, and defenses earned by meeting a security standard. In practice, it applies when you sell or supply software or a connected product, or deploy AI that could injure someone or damage their property or data.
LexLint has not yet researched product and AI liability law in any jurisdiction, so this page lists nothing yet. Until a jurisdiction's product and AI liability law is researched, a lint run there says so in a coverage warning.
Every product and AI liability law instrument LexLint holds, grouped by the jurisdiction that made it. Choose which stages of law to show: the map, the counts and the tables all answer to that choice. To see every area of law, follow Law in the breadcrumb above.
- 1
- 5
- 10
- 20+
- instruments shown
- law on file, none at these stages
- tracked, no law on file
- not tracked
What this page claims, and what it does not
The stages are the in-force tag's. Each filter is one of the
states the tag names across lexlint.io, in its words (in force, not yet in
force, blocked by a court, proposed, no longer in force, and never in force),
and each filter's label wears that tag's mark. In the tables a law in force
is unmarked, its date in plain text, and every other law wears the same tag
the rest of lexlint.io draws: highlighted, boxed, dotted, or struck with its
reason underlined. A withdrawn bill has a filter of its own, never in force:
the /law/<jurisdiction> docket leaves it out, and "no
longer in force" would be false of a law that never was.
Blocked by a court has its own filter. A law a court has blocked is neither in force nor gone: an injunction can be lifted on appeal, and the duty with it. Pooled with a repeal it would hide a count that reads as a signal, how contested a jurisdictional hook is. It is off by default, with the proposals and the laws that have ended.
One date column, and it says what kind of date it is. The date is the day the law started or starts to bind: the earlier of a published start date and the instrument's own date. Neither source is reliably the start, and the earlier one cannot make a law look newer than it is. The column draws the in-force tag, so a bare date never stands beside a law that has stopped: a law in force reads "since" its date, a law not yet in force is highlighted with its date and how far off it is, a law a court has blocked is boxed in red, and an ended law has its dates struck and its reason underlined. Nothing is inferred from our own review date.
The map is the table. Its fill is the count of instruments at the
stages you chose, binned on the same RAMP_BINS edges
/law/map uses, and it redraws on
every change. Point at, click, or tab to a jurisdiction for what applies there,
every stage it holds shown even when your filter is excluding it.
Three ways of showing nothing, and they are different claims. A jurisdiction washed pale holds product and AI liability law your filter is excluding, so widening the filter brings it back. A hatched jurisdiction is one we track and hold no product and AI liability law for at all: that is a statement about our research, not about the law. A white shape has no jurisdiction record for this topic at all. Only the first of the three moves when you change the filter; the other two are facts about us and hold still.
Nothing here needs the script. The default resultset is in the
HTML: counts, tally, map fills and the hidden rows are all rendered at
build. The script recomputes them when a box changes and does nothing else,
which is what lets this page ship under script-src 'self' and
be read whole by a crawler.
Every instrument here links its own note page, which carries the primary source and the date it was read. This is a research index, not legal advice.