AI Slop Session (Bitcoin Core Dev Tech, May 2026)

A discussion of the growing problem of AI/LLM-generated pull requests — roughly 15–20 per week and rising — that are increasingly hard to identify and that drain the project's scarcest resource, reviewer attention. Participants drew a line between bad-faith "remote prompting" (someone acting as a middleman to an LLM) and genuine new contributors, including those leaning on LLMs because of a language barrier, who deserve helpful feedback.

No firm policy was settled. Remedies floated included aggressively closing larger "slop" PRs while tolerating small ones, a disclosure policy, holding every contributor responsible for every line they submit (echoing a "don't paste LLM output" norm), better contributing docs paired with a canned response, and a vouch-style gate before a first PR. The session captured the central tension plainly: the cost of false positives weighed against the time drain on the project, with the requirement that contributors have read and understood every line they post.