Urgent.News

What's breaking now, across thousands of outlets.

More in AI

9 Bugs That All Looked Like a Working System

AgentSelfEdit is an open-source sidecar that rewrites its own system prompt from execution feedback. It A/B tests edits and promotes only statistically-proven winners.

  • Promotion gate incorrectly checked for p-value above 0.95 instead of below 0.05
  • A/B test compared edited prompt against original prompt, not against unedited version
  • Scoring mode marked any trace with non-empty response as 100% correct

More from Tuesday 1 September →