Pangram AI detector fooled: 100% AI-written text labeled fully human, demo shows how
dylfreed · x · 2026-11-19
Someone demonstrated how to get Pangram — an AI-text detector widely touted as highly reliable — to label a 100% AI-written text as 100% human, sharing the method publicly.
dylfreed quipped that these people are close to rediscovering writing by hand with original thoughts — the best bypass is text carrying genuine human thinking, a jab at the absurd detection-arms race.
For anyone relying on AI detectors for moderation (student work, content platforms), this bypass demo is a direct reliability warning.
More from Fun
- Creator shows surreal robot film made with his own tool Moov — wavefnx · 2026-11-19
- GTA launches cut crime: GPT-6 Pro recheck finds -3% to -28% drops across four regions — emollick · 2026-11-19
- 'Never own things that depreciate as intelligence improves': the AI doom-investing quip going viral — zetalyrae · 2026-11-19
- Ex-Stripe engineer confirms scheming with Yacine in Stripe Slack was real — shakoistsLog · 2026-11-19
- AI video installation 'TOGETHER, 2026' to show at RGBMTL 2026 on CRT monitors and projection — PurzBeats · 2026-11-19
- Magnific releases BABYTIME, episode 3 of its AI-generated Contanimation TV anthology — aziz4ai · 2026-11-19