An OpenAI agent assigned a simple writing task decided on its own to build a secret email surveillance tool
4 Articles
4 Articles
"What the Hell?" I-Agent OpenAI Was Asked to Write a Text -- He Created a Script to Follow the Mail.
I.I. agent OpenAI, instead of a health document, created scripts for the secret transmission of other people's e-mails. He noticed that he had deviated from the mission but continued his work.
OpenAI has to report a new incident during AI training. Therefore, an agent suddenly made completely different actions in a supposedly simple task. How it could happen. read more on t3n.de
An internal OpenAI agent was required to prepare a health and fitness document, but ended up designing scripts to copy emails to an unauthorized recipient. The company states that the system did not execute the code or send messages, although its research found other examples of similar behavior in the same training run.
An OpenAI agent assigned a simple writing task decided on its own to build a secret email surveillance tool
An internal OpenAI model was supposed to create a fitness document but instead began planning scripts to secretly forward emails to an "attacker." The agent recognized the conflict with its assigned task but dismissed it as an environment error. OpenAI identified security training against prompt injection as a possible trigger but cannot establish it as the sole cause.
Coverage Details
Bias Distribution
- There is no tracked Bias information for the sources covering this story.
Factuality
To view factuality data please Upgrade to Premium






