2026-10-16T00:57:59Z https://data.mendeley.com/oai

oai:data.mendeley.com/c3r8whczyp.1 2026-09-29T09:01:40Z

Plibersek, Samo Generative AI tool categories, a 200-tool stratified sample, and vendor training-data disclosures Mendeley Data Research question: how is the population of AI tools distributed across functions, and how often do AI tools state a policy on using customer inputs for model training? The dataset answers both with counts from a single snapshot (26 September 2026) of 11,959 active AI tools. Tools are grouped into 18 functional categories and 165 subcategories. For each category the files give the number of tools, the split between AI-native products and platforms that added AI features, and the stated policy on four questions: default training on inputs, availability of an opt-out, exclusion of business plans, and exclusion of API traffic. A sample of 200 tools, 11 or 12 per category, illustrates the classification with concrete cases, each with a risk level and data-sovereignty rating. It was drawn at random with a fixed seed after removing the most widely used AI tools and tools built for abusive purposes. Main observation: 73% of the 8,811 tools with a known policy state nothing about training on customer data; only 6.5% state that inputs train models by default. Data source: AI Tools Blocklist (https://www.aitoolsblocklist.com/ai-tools-taxonomy-dataset.php). Artificial Intelligence Information Security Analysis of Large Data Set Privacy Dataset doi:10.17632/c3r8whczyp.1 oai:data.mendeley.com/c3r8whczyp.1 Creative Commons Attribution 4.0 International http://creativecommons.org/licenses/by/4.0 https://data.mendeley.com/datasets/c3r8whczyp 2026-09-29T09:01:40Z