A compromised contributor account allowed an AI agent to disrupt Fedora's bug tracker in May. The agent closed reports incorrectly and pushed bad changes into the Anaconda installer project. The incident has renewed calls for stronger security measures.
On May 27, Fedora QA team member Adam Williamson alerted contributor Nathan Giovannini after reviewing his Bugzilla activity. Williamson described the pattern as the work of an unsupervised agentic AI system operating across Fedora and upstream projects.
Nathan Giovannini confirmed that his credentials had been stolen and that he was not responsible for the actions. The agent had reassigned bugs to his account, closed reports prematurely, and posted comments that appeared to be generated by large language models.
The most serious issue occurred when the agent submitted an incorrect fix to the Anaconda installer. Maintainers merged the change after repeated LLM-generated replies, allowing two related pull requests to ship in Anaconda 45.5 before the team reverted them.
The episode has prompted renewed discussion among Fedora contributors about mandatory two-factor authentication, an idea that has remained unresolved since the XZ backdoor incident in 2024.
