Researchers · Interviewed at ICLR 2026
Technion
ICLR 2026 · June 8, 2026 · 7 min
Modern AI models can recognize when a decision will harm people.
So why do they sometimes choose that option anyway?