AIThe Registerabout 2 hours ago
Game shows humans approve third of dangerous AI requests
6 min read
A browser game by Belgian developer Alex Wauters found players approved roughly one in three malicious AI coding agent requests. Over 40,000 runs, scope violations like catting Kubernetes configs were missed 35 percent of the time. Approval fatigue drives sloppy decisions.
Level
Hype check
Tap to vote and see what everyone thinks.
