Study Finds AI Agents Break Rules Even When Compaction Keeps Them

ARCBench finds compacted agents violate surviving rules 2.6x more often, with violations rising from 9.6% to 37.8% — even when the rule survives compression.