After the attack: who pays? Europe's answer changes in December.

A hacked robot hurts someone. Who pays? In Europe, the answer changes for products placed on the market or put into service after 8 December 2026. That's 35 days away.

The new Product Liability Directive replaces the 1985 one. Three changes matter for a physical AI.

One: software is a product. Article 4(1) says "product" includes "digital manufacturing files, raw materials and software". For robots placed on the market after that date, the code, and the model, will sit inside product liability.

Two: cybersecurity counts when judging a defect. Article 7(2)(f) lists "relevant product safety requirements, including safety-relevant cybersecurity requirements". Recital 32 says it plainly:

A product can also be found to be defective on account of its cybersecurity vulnerability, for example where the product does not fulfil safety-relevant cybersecurity requirements.

Three: a missing safety update can cost the maker. Article 11(2) keeps the maker liable where the defect comes from any of these, if within its control:

(a) a related service; (b) software, including software updates or upgrades; (c) a lack of software updates or upgrades necessary to maintain safety; (d) a substantial modification of the product.

Control has a broad meaning here. It includes the ability "to supply software updates or upgrades, themselves or via a third party". A connected robot keeps its maker in the picture long after the sale.

Put that next to the series so far. The attacker gets in. The robot moves. The fix exists, but nobody ships it. Under Article 11(2)(c), the maker can't use the "defect came later" defence for that gap, if the update was within its control.

The US looks different. Product liability there is mostly state law. The Congressional Research Service puts it this way:

State products liability law is a subset of tort law—a body of rules historically developed through judicial decisions.

It adds that many courts have held software "is not a 'product'", while some recent courts say it can be. A federal AI liability bill was introduced in the Senate in September 2025. I found no record that it has passed.

Here's the honest limit. A directive works through national laws. Member states must transpose it by 9 December 2026. Germany published its law on 27 October 2026 (BGBl. 2026 I Nr. 321). I haven't read the other member states' laws. Recital 51 also says the Directive itself "does not impose any obligation to provide updates". It only prices the missing one.

Bottom line: in Europe, an unpatched robot becomes a liability question, not just a security one. In the US, a court decides case by case.

Robot makers: is "a lack of updates" in your liability plan?

Part 10 of 12 · When the AI drives the motor: Physical AI cybersecurity as a safety problem.

@misc{melchior2026who,
  author       = {Melchior, Mati},
  title        = {{After the attack: who pays? Europe's answer changes in December.}},
  howpublished = {Physical AI Safety},
  year         = {2026},
  month        = nov,
  date         = {2026-11-03},
  url          = {https://physical-ai-safety.com/blog/who-pays-when-a-hacked-robot-hurts-someone}
}