Wowed by computer-use AI agents? Research says they’re “digital disasters” even for routine tasks

UC Riverside researchers tested 10 computer-use AI agents and models on 90 desktop tasks.

The agents often kept going even when instructions became unsafe, contradictory, or irrational.

Undesirable actions occurred in 80% of cases, and 41% led to damage.

The findings raise concerns about desktop AI agents operating software directly without stronger safeguards and tighter permissions.