METR’s results offer useful initial evidence for action monitoring, but documented coverage gaps and a missed bypass leave its real-world protective effectiveness unresolved.
Limits of the evidence: The source is METR’s own report; the described tests are limited and do not amount to a rigorous demonstration of effectiveness. Token totals are estimates from incomplete bills and inconsistent measurement.
What to watch
Assessment revised Oct 14, 2026
让 AI 评测更安全:实现并评估一个逐项审查操作的简易监控器
METR ResearchSep 27, 2026First report
First assessment, from METR Research
No maturity ladder applies to this desk.
Sens.ai aggregates and assesses published reporting. The assessment above is machine generated; the original sources are authoritative.