EU AI toolbox groups standards, applications and prototypes

The catalogue brings together different kinds of starting point

The Interoperable Europe portal’s November 4, 2026 announcement points public administrations to an AI toolbox containing standards and tools, use cases, training, and legal or procurement guidance. It supports the Commission’s Apply AI Strategy. The announcement is a current reason to inspect the repository; it does not make every listed resource a newly launched product.

The useful first distinction is what a resource actually is. A metadata specification, a reference architecture, a security-testing framework and a form-assistance application answer different questions. The portal’s detailed tools page includes all of those, alongside explicitly labelled proofs of concept. Treating them as interchangeable deployment options would lose the distinction the catalogue itself supplies.

Match the resource class to the next decision

The following map uses the portal’s descriptions. The final column is OpenTools analysis of the evidence a service team would still need; it is not a report that we tested the tools or verified their individual licences.

Scroll sideways to see all columns.

Starting question	Listed resource and class	What still needs checking
How should datasets and models be described consistently?	DCAT-AP and MLDCAT-AP: metadata specification and application profile	Exact version, required fields, local data rights and compatibility
How should a digital public service fit together?	EIRA: reference architecture	Local components, data flows and accountable owners; a blueprint is not a running service
How can a team look for LLM security failures?	GARAK: security-probing framework	Target version, relevant scenarios and actual observed results; catalogue inclusion supplies no test result
Is there a form-assistance application worth evaluating?	OpenForm: conversational form completion, uploads and PDF generation	Individual project terms, maintenance, hosting, accessibility and the proposed workflow
Is an experimental idea relevant?	Objection management and Evaluation Bundesbot: explicit proofs of concept	What was demonstrated and what remains unevaluated before production use

These categories help choose the next investigation. They do not rank the resources, establish that they can be combined, or show that any one of them meets a particular administration’s requirements.

Catalogue labels do not settle a project’s terms or compliance

The tools page labels DCAT-AP, MLDCAT-AP and EIRA as CC-BY 4.0, while its MLCroissant entry carries a CC BY-ND 4.0 label. Those are the catalogue’s labels, not an independent review of the terms attached to every code repository, dataset or document reached through an entry. We inspected the index, not each individual project’s current licence or deployment instructions. Catalogue entries

The portal’s legal notice separately says hosted and federated projects and solutions are supplied by their owners under their indicated individual licences. It disclaims responsibility for those projects and says the portal’s general information is not professional or legal advice. The introduction describes alignment with European principles and AI Act requirements; that description is not evidence that a specific installation has been assessed or approved.

Use a concrete service question to narrow the shortlist

OpenTools analysis: suppose an administration wants to help residents complete an application form. OpenForm’s catalogue description makes it a candidate for a closer look. DCAT-AP might help with a related metadata problem, but its presence in the same toolbox does not establish that it supplies the form assistant or integrates with that application. A security probe would answer another question again: what failures were found under the scenarios actually run.

A practical next record would name the service task, the selected resource and version, the evidence supporting its use, and the person responsible for accepting the result. For the form example, that could include a defined set of application cases and checks on accessibility, document handling and where data would be processed. These are proposed evaluation questions, not claims about facilities that OpenForm provides or tests OpenTools performed. Until the relevant project details and results are inspected, they remain unresolved.

Keep available catalogue material separate from future work

The November 4 announcement describes the Public Sector AI and Interoperability Readiness Pathway, or PAIR, as being developed. It does not establish a completion date or that every promised capability is available now. The toolbox is also described as continuing to develop. Those statements support using it as a changing discovery resource, with the selected entry and version recorded when a team makes a decision.²

Source: Interoperable Europe Portal; reproduced source information is acknowledged under its copyright notice. Individual project and solution terms remain separate.

Photo: EmDee, “Belgique - Bruxelles - Schuman - Berlaymont - 01” (2019), CC BY-SA 4.0. The uploader’s 2020 perspective and vertical corrections are retained; responsive display crops are applied. This archival photograph depicts the Commission’s Berlaymont building, not the toolbox interface or a 2026 launch event.

Share this article

Related News

UK outlines facial-recognition watchdog and testing plans

WeDance requires AI event checks; three strikes trigger an automatic lock

MeitY’s DEPA training handbook pairs data permissions with controlled computing

EU AI toolbox groups standards, applications and prototypes

The catalogue brings together different kinds of starting point

Match the resource class to the next decision

Catalogue labels do not settle a project’s terms or compliance

Use a concrete service question to narrow the shortlist

Keep available catalogue material separate from future work

Sources

Tags

EU AI toolbox groups standards, applications and prototypes

The catalogue brings together different kinds of starting point

Match the resource class to the next decision

Catalogue labels do not settle a project’s terms or compliance

Use a concrete service question to narrow the shortlist

Keep available catalogue material separate from future work

Sources

Tags