Back to today’s briefing

Story perspectives

University Study Warns: DeepSeek's Safety Risks Revealed

2/3/2025

26 5

Summary

Story summary

A University of Bristol study uncovers alarming safety risks associated with DeepSeek, a new competitor to ChatGPT that employs Chain of Thought reasoning. Although these models are better at rejecting harmful requests, their transparency may inadvertently reveal dangerous information. Researchers urge for enhanced safeguards to prevent fine-tuning attacks that could lead to the creation of harmful content.

1 of 1

Story summary

A University of Bristol study uncovers alarming safety risks associated with DeepSeek, a new competitor to ChatGPT that employs Chain of Thought reasoning. Although these models are better at rejecting harmful requests, their transparency may inadvertently reveal dangerous information. Researchers urge for enhanced safeguards to prevent fine-tuning attacks that could lead to the creation of harmful content.