How Long Do Trending AI Projects Stay Available and Maintained? Survival of 733 GitHub and Hugging Face Projects Picked by a Weekly Radar (June–October 2026)

A public weekly AI radar picked 757 projects from 11 June to 31 October 2026; 733 GitHub and Hugging Face projects were observed in 50 daily snapshots (12 September to 31 October 2026). The study asked how many picked projects become unavailable, archived or abandoned, and whether this depends on source, pick month and stars at pick. The metric was pre-registered on 5 October 2026, before analysis. Unavailability (HTTP 404/451 on GitHub, 401/404 on Hugging Face, two snapshots in a row) gave Kaplan–Meier survival of 96.7 % at week 4, 94.6 % at week 8 and 93.0 % at week 12. In total, 23 projects became unavailable, and 5 of them later came back. The composite also counts abandonment (no push for 90 days) and archiving: 90 events, 67 of them abandonment. It falls from 91.9 % at week 12 to 61.5 % at week 16. This drop is produced by the 90-day rule and is not evidence of projects disappearing. GitHub projects in the top star quartile at pick (4,988 stars or more) had 1 event in 144. The difference holds when stratified by pick month (log-rank p = 0.004), but stars and the radar's changing selection cannot be fully separated. No difference between Hugging Face and GitHub was shown (p = 0.21 for unavailability).

For someone citing or building on a project found through a trending list, roughly 1 in 14 projects was gone 12 weeks after the pick (93.0 % available), and 5 of the 23 disappearances were temporary. Abandonment is the more common outcome than deletion: 67 projects crossed 90 days without a push, versus 23 that became unavailable. Low-star picks carried almost all events. The top quartile almost never left within the window, but only 5 top-quartile projects had 16 weeks of follow-up. The data describe one radar and one selection rule over 20 weeks, not GitHub or Hugging Face in general. Statuses before 12 September 2026 are unknown. In practice, when citing fast-moving AI projects, a link to a DOI or an archived copy is safer than a link to a repository alone.