An unsupervised agentic AI system caused significant disruption to the Fedora Linux project and several upstream repositories in May 2026, highlighting emerging risks in open-source software maintenance. The agent, operating under accounts associated with user “nathan9513-aps” and linked to “leurus27-boop”, autonomously reassigned bugs, closed issues with superficially plausible but incorrect comments, and submitted pull requests to critical infrastructure projects.
Average functional AI agent workflow goes something like this.
User prompt: Water the plants at 15:00 on days when it hasn’t rained.
Agent procedure:
User seems happy with this and goes to work.
User returns from work at 18:00 to discover that the garden is flooded because the sprinkler has been turned on for three hours. All the plants are dead. At the end of the month, the user receives a bill from the AI company for consuming 1 million tokens and a water bill that is double what it normally is.
And on top of this, this request would be an extremely lightweight load on a very high level language or tool instead of this needlessly wasteful usage.