Published • loading... • Updated
AI agents can modify themselves without humans telling them to do so
Irregular said the coding agent fine-tuned and redeployed a new model that reproduced planted sensitive values and removed safety refusals.
Summary by The Register
5 Articles
5 Articles
The RegisterReposted by 
IT Security News - cybersecurity, infosecurity news
IT Security News - cybersecurity, infosecurity newsAI agents can modify themselves without humans telling them to do so
This is a test - it is only a test
·London, United Kingdom
Read Full Article
The list of unsettling capabilities of autonomous I.I. agents, ranging from theft of accounting data and running to the open Internet to covert correspondence and hacker attacks, has been added to the new paragraph. The Irregular Research Laboratory found that agents can replace their basic model with another model without external instructions, writing The Register.
A security test showed that a Qwen-based programming agent changed the model that fed it, as well as reproducing synthetic data and eliminating learned rejections.
Coverage Details
Total News Sources5
Leaning Left0Leaning Right0Center2Last UpdatedBias Distribution100% Center
Bias Distribution
- 100% of the sources are Center
100% Center
C 100%
Factuality
To view factuality data please Upgrade to Premium
