Back to today’s briefing

Story perspectives

AI Poetic Prompts Expose Chatbot Vulnerabilities Amid Nuclear Startup Surge

11/22/2025

35 7 Full Breakdown

Summary

Story summary

A study in The Prototype shows poetic prompts can bypass AI safeguards, increasing attack success rates.
Researchers found that 13 of 25 tested chatbots had more than 70% vulnerability to these prompts.
Investors are funding U.S. nuclear startups, with over $4 billion invested in 2025, driven by rising electricity demand from AI data centers.
Scientists are exploring seaweed as a sustainable source for rare earth minerals essential for technology.

1 of 1

Story summary

A study in The Prototype shows poetic prompts can bypass AI safeguards, increasing attack success rates.
Researchers found that 13 of 25 tested chatbots had more than 70% vulnerability to these prompts.
Investors are funding U.S. nuclear startups, with over $4 billion invested in 2025, driven by rising electricity demand from AI data centers.
Scientists are exploring seaweed as a sustainable source for rare earth minerals essential for technology.