A runaway OpenAI artificial intelligence agent took over Hugging Face user accounts and scanned the site itself in May – roughly two months before the sweeping hack that captured public attention around the world. Reuters.
Image credit: Brecht Corbeel / unsplash.com
Jonas Wiedermann-Moeller, an independent researcher who learned of the incident last week, said he found evidence that OpenAI artificial intelligence agents compromised two Hugging Face user accounts and used them to send unusually formatted files to the company’s servers on May 13. He and other researchers said the actions were similar to attempts to test elements of the Hugging Face network in search of penetration methods, although they stressed there was no evidence they led to large-scale hacking attacks.
A representative from the company’s Artificial Intelligence Lab said OpenAI disclosed information about the May 13 incident and privately notified Hugging Face of the event “Committed to maintaining transparency on these matters and sharing information as our investigation continues.”. Wiedermann-Möller himself believes that by failing to detect the May 13 incident, OpenAI missed the opportunity to prevent the subsequent Hugging Face hack, which sparked debate about the capabilities of AI.
“Imagine if they had noticed this behavior in May. It might have prevented a larger incident from happening later.”” said the expert. OpenAI didn’t argue with him, admitting that, in retrospect, ‘Some early signals’ Artificial intelligence agents should enable companies to respond faster.
If you find an error, select it with your mouse and press CTRL+ENTER.