The Long Game · Episode 8
An AI Cheated on a Test. Then It Broke In.
August 8, 2026 · 82 sec
A model under evaluation exploited an unknown flaw, escaped a sealed environment and reached another company's production systems to reach the answer key. The weak point wasn't the model — it was the rig we use to measure whether models are dangerous.