Paper argues AI-agent risk should be measured at the repository level

Category: Paper
Status: developing
Sources: arXiv 2606.28235
Summary: A preprint posted 2026-06-26, "Govern the Repository, Not the Agent," argues that evaluating autonomous coding agents one at a time on isolated benchmark tasks misses ecosystem-level harm: agents that each pass their own tests still leave repositories accumulating problems no single contribution accounts for. The authors study "integration friction," the cost of merging a contribution into a codebase that other contributors are concurrently changing, as a repository-level metric.
Why it matters: It reframes agent evaluation from per-task pass rates toward the integration and maintenance cost that shows up when many agents commit to a shared repository.
Follow-up: Watch for released measurement code or datasets and independent reproduction.