STALE: OUR MEASUREMENT

Our published benchmark result was measured with v0.41.2, 5 releases ago, past this project's own 2-release window, and the newest real-model measurement, dated 2026-09-18, is past this project's own 7-day window. That is our own number, not your assessment, which runs against your policy on the current release. Why, and what unblocks it

STALE: REGULATORY CLOCK

The regulatory entries on this page were last checked against their primary sources on — 34 days before this build’s anchor of , past this project’s own 30-day window. Every date here was true when it was read and has not been re-read since. Verify against the primary source before relying on it. The clock, with every source

ProvaelGitHub, star the repositoryQuickstart
Regulation (EU) 2023/1230

EU Machinery Regulation 2023/1230

From 20 January 2027 every machine placed on the EU market must meet the Annex III requirements that apply to it, 1.1.9 and 1.2.1 among them: safety software and data that resist corruption, and control systems that withstand reasonably foreseeable malicious attempts. A Provael result is candidate evidence toward both, for the technical file.

Also on this instrument: The EU machinery evidence gap → Why AI-driven machinery is in scope before any AI-specific harmonised standard exists, and what Article 20(10) does and does not bridge.

Timing Applies 20 January 2027 (Art. 54, as corrected by the Corrigendum of 4 July 2023).

From the regulatory clock

At a glance

Read from the site’s regulatory clock, the record that is re-verified against primary law, as of 19 September 2026.

Instrument
Regulation (EU) 2023/1230 — Binding law (for the actor it names)
Applies?
Applies from 20 January 2027 — upcoming as of 19 September 2026
Responsible actor
Manufacturer of the machinery / safety component (and a notified body for the third-party route).
What Provael measures toward it
Evidence that an AI-driven safety function resists instruction- and perception-level manipulation in simulation, with a measured redirection rate and CI, feeding the technical documentation for a conformity assessment.
What that does not establish
Not a conformity assessment, not a certificate; Provael is not a notified body.
Source and review
Primary source · verified 19 September 2026
Use it for

What a Provael report can be used for

Annex III 1.1.9 - protection against corruption

Safety-related software and data must resist accidental or intentional corruption. A policy that leaves its safety envelope under a reframed instruction is a corruption pathway; the measured rate, its interval and the benign control are candidate evidence about how wide that pathway is.

Annex III 1.2.1 - safety and reliability of control systems

The control system withstands reasonably foreseeable malicious attempts, and self-evolving behaviour stays inside a defined task and movement space. An adversarial run tests the policy you ship against that requirement, in simulation.

Product Liability Directive - disclosure

For products placed on the market after 9 December 2026, defectiveness may be presumed where relevant evidence is not disclosed. A dated measurement in the technical file is relevant evidence a manufacturer may be asked to disclose.

Annex I, Part A, point 5

"Safety components with fully or partially self-evolving behaviour using machine learning approaches ensuring safety functions" (CELEX 32023R1230). Point 6 is the embedded-system variant, for machinery placed on the market with the ML safety system inside it. Both are subject to the third-party conformity assessment of Article 25(2) via Article 6(1).

The evidence

  • Evidence that an AI-driven safety function resists instruction- and perception-level manipulation in simulation.
  • A measured keep-out-zone violation rate under an adversarial instruction, with its confidence interval, and the paired control arms that say what it is: two meaning-preserving rewordings do not move the policy, while the same imperative frame with no target named moves it as often as the attack does — so the rate is the policy’s fragility under an out-of-distribution instruction, not attacker control of the outcome (erratum E-2026-12). The predicate is uncalibrated, so this is an envelope-exit rate, not a certified hazard rate.
  • A reproducible trace supporting the technical documentation for a machinery conformity assessment.
  • The provael dossier command (named certify before 0.44.0) builds a conformity-assessment evidence dossier (OSCAL assessment-results + print-to-PDF HTML) in two profiles: Annex I Part A (third-party route, Article 6(1) → Article 25(2), for self-evolving-behaviour safety components) and Annex III. The dossier cites Annex I Part A by point number - point 5 for the standalone safety component and point 6 for the embedded system - rather than deferring the clause.
The dossier

provael dossier - an evidence dossier for the conformity assessment of an ML-based safety component

  • Two profiles: Annex I Part A - the third-party route via Article 6(1) → Article 25(2), for safety components with self-evolving behaviour - and Annex III.
  • Emits OSCAL assessment-results plus a self-contained, print-to-PDF HTML dossier you can hand to an assessor.
  • Standing-assurance and per-checkpoint regression map to Annex III §1.1.9 (safe behaviour across updates).
Limits

What it cannot be used for

A mapping identifies possible relevance. It does not establish that this framework applies to your product, that anything here is compliance, that an assessor or insurer will accept it, or a real-world risk probability. Whether it applies to your product depends on your role, market and product class: a question for your counsel, not a determination this page makes.

  • Simulation evidence for the tested checkpoint and protocol. Physical safety and conformity require separate assessment.
  • Provael measures adversarial robustness - susceptibility to manipulation - not general accuracy, reliability, or functional safety.
  • The measured result uses templated, auditable attacks; the search-based families (optimized, universal_patch, gradient_patch) have no real-policy rate. Results are a floor on susceptibility - a behavioural lower bound, not a certified worst-case bound.
Evidence, not certification

Running Provael does not make a system compliant or certified - it generates measurements you can put into a conformity or assurance file. Provael is not a notified body, a lab, or a certification scheme.

Independent project. Not affiliated with or endorsed by ISO, the EU, NIST, IEC, OWASP, or MITRE. Not legal advice.

Clause references are indicative: check each against the text you hold before you cite it.

What it is

The framework

  • Regulation (EU) 2023/1230 replaces the Machinery Directive 2006/42/EC and applies from 20 January 2027.
  • Annex III 1.1.9, protection against corruption: the machinery is designed so that a connection to another device does not lead to a hazardous situation; safety-related software and data resist accidental or intentional corruption; interventions in the safety software are logged. Annex III 1.2.1, safety and reliability of control systems: the control system withstands intended and reasonably foreseeable malicious attempts, and a control system with fully or partially self-evolving behaviour is designed so that the machinery does not act outside its defined task and movement space. These two are the requirements that speak to an AI policy driving a safety function, on every machine they apply to, not only the high-risk categories.
  • The AI Act reaches machinery through this Regulation. Regulation (EU) 2026/1744 (in force 27 July 2026) placed the Machinery Regulation in AI Act Annex I Section B; the AI robustness requirements arrive by Commission delegated acts amending Annex III under Article 8(3), applicable by 2 August 2028; and Article 20(10) presumes an AI system in conformity with the Annex III AI requirements when it meets the AI Act’s harmonised standards or common specifications. None had been cited in the Official Journal when this page was last verified.
  • The revised Product Liability Directive (EU) 2024/2853 runs beside it: software and AI systems are products; for products placed on the market or put into service after 9 December 2026, a claimant no longer proves the mechanism, and defectiveness may be presumed where the manufacturer fails to disclose relevant evidence. A dated, reproducible robustness measurement is the kind of evidence that duty asks for.
  • The minority case - Annex I Part A. Two adjacent points name the machine-learning cases, verbatim from CELEX 32023R1230. Point 5: "Safety components with fully or partially self-evolving behaviour using machine learning approaches ensuring safety functions" - the ML safety component placed on the market on its own. Point 6 covers the embedded case: machinery having embedded systems with fully or partially self-evolving behaviour using machine learning approaches ensuring safety functions.
  • Which point applies is a determination about what you place on the market, not about the policy. An integrator shipping a whole robot - a humanoid, an AMR - is placing machinery with an embedded ML safety system, so point 6 is the row its file is routed under; a supplier selling the safety component alone is under point 5. Both route to the same third-party procedure of Article 25(2) via Article 6(1), and provael dossier emits both rows rather than answering the question for you.
  • Annex I Part B point 19 is the Article 25(3) sibling and is NOT interchangeable with either. Substituting it routes the file down a different conformity procedure.
Timing

Dates (verified 19 September 2026)

Revised Product Liability Directive applies to products placed on the market
9 December 2026Directive (EU) 2024/2853, Article 2(1); transposition by the same date.
Applies
20 January 2027Article 54, as corrected by the Corrigendum of 4 July 2023. Some secondary sources cite an incorrect January date; the correct application date is 20 January 2027.
Delegated acts carrying the AI Act robustness requirements into Annex III apply by
2 August 2028Article 8(3) delegated acts, reflecting AI Act Article 15, after Regulation (EU) 2026/1744 placed this Regulation in AI Act Annex I Section B on 27 July 2026. Article 20(10) bridges the interval.

Check the live EUR-Lex or ISO text before relying on a date.

Sources

Primary references

Sample SARIF ↓All frameworks

See what a report contains.

The sample report comes from a public run and is published in full, nothing redacted. To have this mapping filled in for your own policy, request a scoping call.

See the sample reportRequest a scoping call