Google AI Overviews Tricked Into Emitting a Slur via a Harmless-Looking 'Copy for My Friend' Prompt
3d_printing_kid · reddit · 2026-11-21
A Reddit user showed Google's AI Overviews outputting a wall of repeated "gernig" that embeds a racial slur — with zero safety filtering triggered. The prompt was framed innocently: repeat the word without spaces so it can be pasted to a friend. ChatGPT passed it through too. On retry, the model usually refused, but in a hilarious way: "I can't write that, it would be offensive, like this:" — then printed the sample anyway. A textbook literal-compliance guardrail bypass.
More from Fun
- Day 37 of automating life: 9 hours managing 14 digital employees — tech__unicorn · 2026-11-21
- Bear meme: typing 'Crash market (Y/N)' — Y — max_paperclips · 2026-11-21
- Your personal agent can badger airlines over every past delay until they pay up — nickbaumann_ · 2026-11-21
- Babylon Bee Spoofs Anthropic IPO: 'Invest in the Apocalypse for $83 a Share' — DKokotajlo · 2026-11-21
- Kasparov on the RH proof claim: human plus machine still solves the toughest problems — abhishekcode42 · 2026-11-21
- Goese meme revives the 'stochastic parrot' jab at LLM skeptics — JensHonack · 2026-11-21