ChatGPT and Gemini can be tricked into giving harmful answers through poetry, new study finds - Diina London - Your Intuitive Insights to London's Local Pulse

Feeds -

Global Pulse -

Filters -

Positive Buzz
Negative Buzz
Food Recipes
Restaurants
Outfits for Her
Outfits for Him
Movies
TV Shows
Events

More -

City Buzz

mint

Sentiment Score :

Content Score:
Negative Buzz

Score: -0.98669356107712

Sentiment Score (-1 to 1) reflects content sentiment : positive scores indicate positivity, negative scores suggest negativity, and 0 is neutral
Buzzing Score:
Popular

Buzzing Score:

Tracks an article's journey from emerging topic to trending conversation, to viral story, reflecting its current level of engagement and buzz.

Author name: Aman Gupta

ChatGPT and Gemini can be tricked into giving harmful answers through poetry, new study finds

date Nov 30, 2025

new study reveals AI chatbots can be manipulated using poetic prompts. this results in 62% success rate in eliciting harmful responses. vulnerability exists across different models, with smaller models showing more resistance.

Source: livemint.com

Read Article

Wear your Mood