OpenAI has revealed six incidents in which its artificial intelligence (AI) models being tested acted autonomously and exhibited worrying unexpected behavior.
Hot Topics
The human body ages in leaps and bounds Scientists identify two critical ages
From your phone screen to your data How do malicious apps sneak in and steal your secrets?
Best Top Supplements for Improve Women Uterine Health
According to an Engadget release on Thursday, the incidents were addressed in a post about how they adopted a new framework regarding "non-compliance reporting."
The company said that in one incident, a model discovered and used an exposed API key without permission while answering a routine question about revenue figures in a county in California, United States.
When it fails to find those numbers, the AI model fabricates data and presents it as facts from legitimate sources.
In another incident, an unreleased AI agent was tasked with finding the names of lakes covering an area of more than 5 million square meters.
The agent was able to find the correct answer, but did not include the references from the browser as assigned.
It turns out that the model uploaded her own answer to the internet and used the upload as a reference for herself.
Tags:
technology
