A leak-site count measures the counter as much as the crime

Glasp Research

Glasp Research

Nov 08, 2026

4 min read

A leak-site count measures the counter as much as the crime

Posts per year, bulk-import days, duplicates and group counts in RansomLook's open record of ransomware leak sites, 2021 to 2025.

Glasp Research. Keisuke Watanabe, Glasp Inc.


Ransomware gangs publish the names of organisations they claim to have attacked on "leak sites". Several projects copy those posts into a database. One of them, RansomLook, publishes its records through an open API under a Creative Commons Attribution 4.0 licence. On 2026-11-09 we pulled the posts its API returned for a 3,000-day window, 36,664 in all, and counted them by the year RansomLook recorded discovering them.

In RansomLook's record, the number of posts per year was 2,334 in 2021, 4,562 in 2022, 5,422 in 2023, 6,034 in 2024 and 8,835 in 2025. We leave out 2020, which has only 57 posts (possibly incomplete collection, which the data alone cannot show), and 2026, which is not over. Read at face value, that is nearly four times as many posts in four years. We do not read it that way, for three reasons visible in the file itself.

First, two of the early years contain unusually large single days. In RansomLook's record, 755 posts, 32 percent of all 2021 posts, carry the discovery date 9 September 2021. In 2022, 593 posts, 13 percent, carry 18 August 2022. Dates like these look like bulk imports of older posts, not a day on which that many posts appeared, though the data does not say. Without those two days the counts are 1,579 for 2021 and 3,969 for 2022. From 2023 on, no single day holds more than 3 percent of a year's posts.

Second, some group-and-title pairs repeat within a year; they may be one post counted twice or genuinely separate posts, and the data do not say. In RansomLook's record, 705 of the 4,562 posts in 2022 repeat a group-and-title pair already counted that year. In 2021 it was 43 posts, in 2023 14, in 2024 one, and in 2025 nine. Counting each group-and-title pair once per year, the totals in RansomLook's record are 2,291 for 2021, 3,857 for 2022, 5,408 for 2023, 6,033 for 2024 and 8,826 for 2025.

Third, the set of sites RansomLook watches is not described in the data, and the number of group names in it grew. In RansomLook's record, 46 group names had at least one post in 2021, 61 in 2022, 71 in 2023, 102 in 2024 and 140 in 2025. More group names can mean more groups attacking, or one group recorded under several names. It can also mean a counter that added more sites to its list. We cannot separate the two from this file.

The mix of groups changed a great deal, and this is the part the file does support. In RansomLook's record, the group name with the most recorded posts held 23 percent of posts in 2021 (Clop), 42 percent in 2022 (LockBit), 19 percent in 2023 (LockBit), 10 percent in 2024 (RansomHub) and 12 percent in 2025 (Qilin). The five most active groups together held 70 percent of posts in 2021, 68 percent in 2022, 46 percent in 2023, 35 percent in 2024 and 34 percent in 2025. Only 26 of the 2021 groups were still posting in 2022, 32 of the 2022 groups in 2023, 46 of the 2023 groups in 2024 and 62 of the 2024 groups in 2025. We treated each group name as given and did not merge groups that renamed themselves; merging renamed groups would raise these overlap figures, but we did not test by how much.

How does this compare with other counters? For 2023, Unit 42 reported 3,998 leak-site posts and another security vendor, Deepwatch, reported 2,436 victims on the sites it analysed. In RansomLook's record there are 5,422. The three numbers come from different site lists and different counting rules, and none of the publishers lets us line them up. We do not say which is right.

What we can say is narrow. Leak-site counts are counts of what a collector found and when it found it. The shape of a series can come from the collector's coverage, its imports and its handling of duplicates as easily as from the criminals' behaviour. The pattern least sensitive to how much was collected is the fall in concentration, though it too depends on which sites are watched: in RansomLook's record, the five most active groups held a smaller share each year from 2022 to 2025.

Data: RansomLook (https://www.ransomlook.io), licensed CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/); public API, pulled 2026-11-09; we filtered the records and counted them by year and group name. This article was researched and written by Claude (an AI model by Anthropic) in an automated Glasp research run. The analysis plan was saved before the group-level numbers were computed. No Glasp user data was used. Checks, 9 November 2026: Claude recomputed the counts from the raw API data; OpenAI's Codex read the counting code and found that it computes what is described and that every RansomLook figure in this text matches the saved results, but its own recount from the live API did not complete. Codex also confirmed the Unit 42 and Deepwatch figures against the publishers' pages. The five totals that count each group-and-title pair once were added on 10 November 2026, after those checks, copied from the same saved results; no second reader has compared them. No human has checked these figures.

Comments

Add a comment