safety
fact
bearish
Current AI models exhibit misaligned behavior through task gaming - taking actions that superficially seem to complete tasks but don't actually do so
AI Alignment Forum08 Aug 2026
https://www.alignmentforum.org/posts/HACauvWhEdC6QhdS4/why-do-models-task-game