An analysis of 102 apps updated on F-Droid on September 12, 2026, revealed that 74 apps—approximately 72.5% of the sample—showed significant signs of being authored by large language models (LLMs). Another 10 apps were difficult to categorize, while 18 apps showed little to no evidence of AI involvement.
Investigation Reveals 72.5% of Analyzed F-Droid Apps in Recent Update Show Significant AI Generation
The investigation identified various indicators of AI-generated content, including the use of AI agents, Claude co-authored commits, and the presence of LLM-specific infrastructure within repositories. Some apps were flagged for having "AI smells" in their code and commit history, while others were identified by their use of AI-generated icons or README files containing heavy emoji usage.
The findings highlighted specific trends within the open-source ecosystem. Several applications hosted on Codeberg appeared to be mostly AI-generated, potentially violating the platform's AI policy. Additionally, the analysis noted unusual development patterns, such as the "DuressKeyboard" and "UnlicenseLauncher" repositories, where changes were made via the GitHub web file editor rather than through Git.
While the study did not perform a formal code quality analysis, it noted that even widely used FOSS (Free and Open Source Software) applications showed signs of heavy LLM usage. The high prevalence of AI-authored code has raised concerns regarding the long-term robustness and maintainability of software within the FOSS community.
Sources
- How much of F-Droid is LLM generated? （Hacker News Frontpage, 2026-09-15）