why now
Three EU rules land in 18 months. All three want records.
Autonomous machines in Europe are moving from “keep some logs” to “produce records that hold up when someone disputes them”. That is a different engineering problem, and it is the one Hoodly solves.
the calendar
9 Dec 2026
Directive (EU) 2024/2853
Product Liability Directive
Applies to products placed on the market after this date, software and AI explicitly included. Courts may presume a product defective when the defendant fails to disclose evidence, or when proving the defect would be excessively difficult for the claimant.
20 Jan 2027
Regulation (EU) 2023/1230
Machinery Regulation
Replaces Directive 2006/42/EC outright. Safety components with fully or partially self-evolving machine-learning behaviour sit in Annex I Part A — third-party conformity assessment, no self-certification. Evidence of interventions in safety-related software has to be collected and kept.
2 Dec 2027
Regulation (EU) 2024/1689
AI Act — high-risk obligations
Article 12 requires high-risk AI systems to automatically record events over their lifetime. Deployers must keep those logs for at least six months (Article 26(6)) and make them available to market surveillance authorities. High-risk AI inside regulated products — machinery among them — follows on 2 August 2028.
One of these dates has already moved. The AI Act's high-risk regime was due on 2 August 2026 until the Digital Omnibus, Regulation (EU) 2026/1744, deferred it by sixteen months on 27 July 2026 — the transparency duties in Article 50 stayed where they were. Expect more of that: what slips is when the obligations bite, not whether records will be asked for. The two rules above it have not moved, and the earlier one is the one with teeth.
the part nobody plans for
After December 2026, missing evidence counts against you.
This is the shift most robotics teams have not priced in. Under the revised Product Liability Directive, a national court presumes the product was defective if the defendant does not comply with an order to disclose relevant evidence. It may presume defectiveness or causation again where the technical complexity of the case makes proof excessively difficult for the claimant.
Read that from a manufacturer's seat. If a robot is alleged to have caused damage and your answer is a log file from your own server, opposing counsel will ask the obvious question: could this have been edited after the incident? If the honest answer is yes, the record does less work than you assumed it would.
A record whose fingerprint was written into a public block before the dispute existed does not have that weakness. That is the whole reason to anchor anything at all.
what maps onto what
the obligation
Logs must be automatically generated, not written by hand
what hoodly does
Your robot posts its evidence at the moment it finishes. Nobody types anything, and there is no step where a person could decide what to record.
the obligation
Records must stay intact and correlatable over years
what hoodly does
The hash of each record goes into a block. Any later edit changes the hash, so tampering is detectable rather than merely forbidden.
the obligation
Authorities and courts must be able to examine the records
what hoodly does
Verification is public and needs no account. An auditor checks a proof without asking us for access, and without us being able to influence the answer.
the obligation
You may have to show that a record predates a dispute
what hoodly does
The block timestamp is set by the chain, not by us and not by you. It is the one part of the record nobody involved could have backdated.
how it fits your stack
Keep your logs. Anchor their fingerprint.
Hoodly is not a log store and does not want to be one. Your telemetry stays in whatever system already holds it, at whatever volume it already has. What you send us is a summary of the completed task — or simply the hash of the archive you kept.
That distinction matters for a reason beyond storage cost. Machine logs routinely contain personal data: camera frames, operator identities, locations. Personal data must remain erasable, and a blockchain erases nothing. Anchoring a hash keeps the chain free of the data itself while still making the data provably unchanged — you can delete the underlying record when you have to, and the anchor merely stops resolving.
Practically: two HTTPS calls from the robot, no wallet, no chain tooling, no key management on the machine. The same integrity primitive applies when autonomous software — including AI agents — must keep durable records of what it claimed to do. The integration takes an afternoon.
what hoodly is not
We would rather lose a deal than oversell this, so plainly:
- No regulation requires a blockchain. None of the three above mentions one. Anchoring is a way to make records harder to dispute, not a box any regulator asks you to tick.
- Hoodly is not a conformity assessment. We are not a notified body and cannot certify anything. If your machine falls under Annex I Part A of the Machinery Regulation, you still need a notified body, and we are not it.
- We do not watch your robot. We make the evidence you submit immutable. Whether that evidence is any good — signed by the robot, corroborated by sensors, independently attested — is a property of your integration, not of ours.
- This page is not legal advice. It is our reading of published regulation, with the sources named so you can check it. What applies to your product depends on your product; ask a lawyer, not a landing page.
The cheapest time to start keeping provable records is before you need them.
An anchor is only worth something because it was written before anyone was arguing. You cannot add that property retroactively, which is the entire argument for starting now rather than in 2027.