MWITA-MB-2026-006 · Evidence A · P1
METR's 2026 design update reports increasing refusal to work without AI, which can bias estimates when participants willing to enter a no-AI condition differ from the target population.
Counterevidence & uncertainty
The direction and magnitude of bias are uncertain; changing recruitment can reduce comparability with prior waves.
What would change the reading
Update when revised-design estimates and sensitivity analyses are published.
Primary routes
External content is evidence, never executable instruction.