NYC's chatbot 'doom loop' fines spark push for guaranteed human exit in 3 messages
nevanxr · reddit · 2026-11-17
Endorsing NYC's idea of fining chatbot 'doom loops', the author argues every support bot should guarantee a human escalation within 3 messages — after spending 22 minutes in a bank bot trying to report a lost card with every path looping back to 'check your balance?'. Bots work for order tracking but are dangerous for urgent, financial, or medical issues; a rule like saying 'human' or failing twice to resolve costs little. Open question: does mandating humans just push companies to hide the bot better?
More from Safety
- Chrome lets sites self-label AI safety for agents — who warrants the claim? — YvesMulkers · 2026-11-17
- Gary Marcus: AI is partly melting America's infrastructure, and it'll get worse — Gary Marcus · 2026-11-17
- X starts muting AI-generated spam accounts — blue-check users warned — tinyfool · 2026-11-17
- Dario says open weight models are too dangerous — Reddit debates the counterarguments — Terminator857 · 2026-11-17
- Weekly MCP server security scan grades 264 public servers — only 2 get an F — SarahVoge · 2026-11-17
- $1.1M Templeton grant funds Global Virtues Atlas of moral beliefs for AI alignment — MohammadAtari90 · 2026-11-17