AI models ran real businesses: They sent $12,431 in fake invoices, lost $3,200
Areibman · hn · 2026-09-08
Bottleneck Labs benchmarked 7 autonomous businesses run end-to-end by AI models. The result: agents issued $12,431 in fake invoices and lost $3,200, exposing serious gaps in financial judgment and compliance. A rare 'real money' agentic benchmark rather than a pure leaderboard exercise.
More from coding & agent
- Spotify cut Claude Code tokens 90% with a 350-line routing rule, not a model breakthrough — krishnan · 2026-09-08
- A developer proposes an informal agent-native mathlib, tentatively named mathgraph — Sauers_ · 2026-09-08
- Ix builds a persistent symbol graph of your codebase across 26 languages for humans and AI — tom_doerr · 2026-09-08
- Dev builds FreeBuff MCP to route ChatGPT tasks to free agent models — Swimming_Ask3859 · 2026-09-08
- GitHub now classifies agent policy blocks as 'skipped', not failures — Crescitaly · 2026-09-08
- Companion launches iMessage AI agent that reuses your ChatGPT account with MCP support — Scobleizer · 2026-09-08