Last week, ChatGPT maker OpenAI released what appeared to be a frightening memo about an “unprecedented cyber incident” in which its cutting-edge AI model went rogue and hacked another company.
AI Lab has revealed that a group of next-generation AI models, including some not yet publicly available, escaped from a secure test environment and entered a production database owned by machine learning startup Hugging Face.
“It’s absolutely amazing that all this happened automatically,” Hugging Face CEO Clement DeLang tweeted, seeming to back up the story. “The investigation is ongoing and we will share further information from what we believe is the first incident of its kind.”
Mainstream news outlets picked up on industry coverage and went along, publishing a flurry of disturbing reports about dangerous AI. one CBS The headline said: “Very worrying” NBC He feared it was “just the beginning” of a creepy trend.
But if you look beyond the alarming headlines, you’ll find a common pattern. That is, big tech companies are using fear tactics to manipulate public opinion.
John Sixtan is an assistant professor of computer science at Cornell University. guardianThis isn’t the first time OpenAI has spread the word about “scary AI.” Back in February 2019, long before ChatGPT was publicly available, the company announced GPT-2, the then-state-of-the-art large-scale language model. But the company argued that GPT-2 was too dangerous to release to an unwise public, and the announcement fed the AI hype that led to OpenAI securing a hefty $1 billion investment from Microsoft.
“This was an early example of OpenAI’s communication pattern,” Sixtun explains. “We will proclaim loudly how dangerous AI is, and investors will learn how powerful it is.”
It remains to be seen whether OpenAI’s rogue agent story will hold up to scrutiny, but there are reasons to be skeptical.
If OpenAI’s model were indeed gnawing at the wires of other American companies, it would pose a serious threat to economic activity, and a worthy Justice Department would want to shut it down (though it might not). this It’s a repeat of the Department of Justice, but you get the idea. )
And as AI Labs does, OpenAI is famously safe from both outside prying eyes and insider threats inside the hall. It’s certainly possible that OpenAI is exaggerating its own internal safety measures as a kind of security theater. But on the flip side, OpenAI is now the primary AI contractor for the U.S. Department of Defense, a relationship that requires cutting-edge security practices.
It similarly follows the broader AI industry strategy. Anthropic, the company behind Claude, has deployed similar rhetoric, warning of a similar outbreak of rogue AI agents breaking out of their cages on their own. In a bizarre example, the company’s billionaire co-founder Chris Oler once warned the Pope that the Vatican needed to help stop rogue AI from “taking over humanity.”
More information about OpenAI: Man sues OpenAI, claiming ChatGPT nearly killed him with horribly dangerous medical advice
