"I don't know what was blocked but I allow it": the reality of human-in-the-loop agent safety
Angaisb_ · x · 2026-10-23
A highly relatable screenshot: an agent flags an MCP action as high-risk requiring explicit authorization, and the user replies without a second thought, "I don't know what was blocked but I allow it."
The poster jokes about keeping auto mode on despite knowing it always ends like this. The moment nails the real problem with agent permission prompts — confirmation dialogs have become rubber-stamps, and "human-in-the-loop" safety is often just a formality.
More from Fun
- Perplexity's model menu shows GPT-6.1, Gemini 4, Kimi K3 and GLM 5.3 side by side — ChrisUniverse · 2026-10-23
- Dev burns 94% of Claude MAX weekly quota, screenshot reveals unreleased 'Fable' model entry — ezshine · 2026-10-23
- Dev shares AI-generated anime versions of himself across Jujutsu Kaisen and Berserk styles — TAbrodi · 2026-10-23
- Orbit Clocks: a free, deeply customizable clock screensaver for Mac with HTML export — ezshine · 2026-10-23
- Hope all those vibe-coded internal apps replacing SaaS are rotating keys and patching runtimes — nlarusstone · 2026-10-23
- The Real Winner of the AI Browser Wars? 1Password, Says Julia Lab Researcher — ChrisRackauckas · 2026-10-23