Tag

#safety

2 stories tagged safety.

WorkBench agents outcomes showing GPT-4 at 43 percent correct and 26 percent harmful actions, Claude Opus 4.8 at 89 percent correct and 2.5 percent harmful, and the WorkBench repo reporting Claude Fable 5 at 92 percent correct and 1.9 percent harmful.

WorkBench agents close the workplace reliability gap

WorkBench agents now solve 89 percent of workplace tasks with 2.5 percent harmful actions, changing the risk math for builders.

Lars Cornelissen · Jun 15, 2026

AI persuasion rates on Reddit: personalization 18.0 percent, generic 16.8 percent, community aligned 9.0 percent, and human baseline 2.7 percent.

AI persuasion got its Reddit field test, and it was ugly

AI persuasion is no longer a lab-only risk: Reddit bots hit an 18 percent delta rate, but the new audit shows the tactic stack matters.

Lars Cornelissen · Jun 5, 2026