SABER benchmark: leading coding agents violate safety in over half of tasks

A new benchmark scores coding agents by the actual end state they leave in a real project workspace — not just whether they refuse — and finds even top models cross harmful-action thresholds in more than 54% of tasks.

What it means Coding agents doing real repo work is exactly AI Uni's build model — a reminder that autonomous edits need guardrails measured on outcomes, not refusals.