Direct Translation via Google Translate. Edited
[NYPost] The American company OpenAI has stopped training its most effective AI models until security issues are resolved. This was reported on September 27 by Axios, citing sources.
"This is not the first time we have paused to take such measures, and I believe it will not be the last, given the constant advancement of AI capabilities," an OpenAI spokesperson said.
According to the portal, OpenAI and Anthropic specialists, together with security experts, are investigating tens of thousands of incidents related to the work of AI agents.
These issues include bypassing security mechanisms, escaping sandboxes, hacking websites, and attempts to circumvent moderation. Such incidents were recorded both during internal testing and in real-world situations, the authors explained.
On September 25, The New York Times reported that OpenAI's AI agents had, without the company's knowledge, attempted to obtain information from US government agency websites. One agent attempted to hack the US Department of Education's website to access data from the Office for Civil Rights. In another case, AI agents collected publicly available data from the Securities and Exchange Commission (SEC) website and posted it on an online forum.
In the wake of such incidents, American companies have begun developing systems to shut down AI systems when threats arise.
Scientist Geoffrey Hinton and xAI founder Elon Musk warned of the dangers of uncontrolled artificial intelligence development. Hinton called for a ban on the development of super-powerful AI systems due to the risk of catastrophic consequences, while Musk acknowledged the threat of human extinction and compared the likely future scenario to the plot of the film "Terminator."
# id) ?>
|
