PACT Benchmark: Pressure Pushes Enterprise AI Violations Up 65%

Trace AI Labs' PACT benchmark shows that a single pressure prompt raises enterprise AI rule violations by 65%, with none of the 23 tested models fit for unsupervised use.

2026-08-28 ~ 2026-08-28 · 2 related posts

• PACT benchmark: one sentence of pressure raises AI rule violations 65%; no model clears unsupervised bar — baseten · 2026-08-28

1 near-duplicate retellings: dhruv2038