Welcome to aisecuritymapping.com – a website for comparing AI security requirements, recommendations and standards from around the world.
The UK Government commissioned Copper Horse to identify and map AI security materials against the European Telecommunications Standards Institute’s European Norm or Standard, ETSI EN 304 223 Securing Artificial Intelligence (SAI); Baseline Cyber Security Requirements for AI Models and Systems. The security requirements were assessed against the 13 principles defined within ETSI EN 304 223, which are grouped into five lifecycle phases: Secure Design, Secure Development, Secure Deployment, Secure Maintenance and Secure End of Life.
Mind maps have been created for each principle to identify and highlight similarities and differences between ETSI EN 304 223 and alternative standards, recommendations, guidance and government publications from around the world. These interactive mind maps capture the text of every mapped requirement, enabling detailed comparison and analysis across the various source documents.
A visualised example of the mindmap for Principle 13 can be seen below:
This website also provides open datasets for all 13 principles in JSON, CSV and ODS formats, alongside downloadable PDF and PNG versions of the associated mind maps. These resources can be accessed by following this link.
Our Approach
Copper Horse carried out open-source research to identify any relevant standards, requirements and recommendations. Initially the UK government provided a set of documents which were later combined with discovered alternative proposals. The researchers then mapped all requirements and recommendations to the 13 principles set out in the European Telecommunications Standards Institute (ETSI), EN 304 223. Every mapping was checked by multiple researchers to verify their most related principle; however, it should be noted that some requirements may come under multiple principles which could result in differing opinions. The main goal of this was to identify how AI security is currently being developed and how that related to what is set out in the EN 304 223. The researched showed that ETSI EN 304 223 covered the vast majority of the 2,182 AI security requirements reviewed in this project.
Documents from countries across the world were reviewed, however, to ensure accurate mappings, only documents originally in English or with official English translations were included. A list of potentially relevant documents captured but with no official translation can be found in the document repository.
We recognise that standards have been created and published by BSI, ISO and CEN/CENELEC which can be accessed on their respective websites. Many of these have not been mapped as they require a user to purchase a licence. However, in some cases the researchers have identified potential topics which these standards might cover allowing for a wider picture of the current state of AI security standards. More information can be found on the FAQ page.
Other documents are currently still being drafted, in this case the draft version has been mapped. It is worth noting that there might be changes to these requirements after the publication of this website.
A table showing the number of mappings to each principle for every document is available to view here.
Submissions of other documentation for future consideration are welcomed at: aisecuritymapping[@]copperhorse.co.uk
A visualised example of a mapping on a mindmap can be seen below:
Each document in the mindmaps have a link to their source. Each requirement node has a note specifying the actual requirement text. Click the chain on a document node to access it’s URL, click the horizontal lines with a “+” on a requirement node to access the note.
