Methodology

A company was included when its category metadata identified it as an AI company or referenced AI technologies, including generative AI, machine learning, deep learning, neural networks, natural language processing, computer vision, large language models, AI infrastructure, and chatbots. Each candidate was then reviewed against its business description and public website to confirm that AI is central to its products or services; companies that only use AI incidentally were excluded.

AI activity was determined through business and technology classifications and a review of company products and services. Activities are presented in six broad groups, with a small unclassified category.

Europe is defined geographically and includes the EU-27, the United Kingdom, Switzerland, Norway, and the rest of the continent, excluding Russia and Türkiye. Taiwan is considered separate from China. Company-level findings are reported as percentages, with each company counted once per category. Country and regional percentages refer to the companies within the relevant geographic group; different security categories may overlap.

used to build the study population; size is available for 86% of companies, founding year for 88%, and disclosed funding for 26%.

632 manufacturing vendors, 735 healthcare vendors, 2,690 IT vendors, and 10,000 finance vendors.

Critical vulnerabilities are identified by matching the observed versions of internet-facing software, operating systems, and network devices against publicly disclosed, critical-severity CVEs. Credential findings include company services and third-party business applications. All assessments were conducted externally, without access to company systems, using only publicly observable information. Unusual network behavior is identified from NetFlow traffic records compared against each company's own historical baseline. Every vulnerability reported here is a potential exposure identified from the outside; the research did not test exploitation, and the findings are not, by themselves, confirmation of a successful attack.

MCP severity percentages describe recorded findings rather than companies. Domain attribution does not establish production deployment, and private MCP environments are outside the analysis. Forbes AI 50 results describe a separate supplementary cohort.

PHP2Shell figures come from a supplementary scan completed in the week after the October 21, 2026 fixes and are based on observed PHP versions on internet-facing systems.