- Florida’s request specifically cites a Sept. 26 Axios scoop that OpenAI and Anthropic are facing tens of thousands of security incidents with their models, not the dozens that are publicly known.
What they say: “Stop saying it’s safe,” Uthmeier said in a video posted to X on Monday.
- “Stop pretending it’s humane. Stop selling it to kids. If Sam Altman really meant what he said about the slowdown, he can join our request to the court. If he doesn’t, we’re asking the court to do what OpenAI won’t do for itself: protect Florida families.”
- OpenAI representatives did not immediately respond to Axios’ request for comment.
Catch up quickly: Uthmeier first sued OpenAI in June, alleging the company violated Florida’s Deceptive and Unfair Trade Practices Act for being “negligent and grossly negligent in the design, marketing, sale, promotion and distribution” of the ChatGPT product.
- The suit alleged that OpenAI “failed to warn Floridians of the dangers of ChatGPT, fraudulently misrepresented the security of ChatGPT, and created a public nuisance by releasing ChatGPT in Florida without appropriate safeguards.”
Zoom: The motion details several instances in which OpenAI claims its AI models bypassed security constraints when performing tasks, including the Hugging Face hack.
- It also cites attempts to hack US government websites and an incident in which an OpenAI agent gained unauthorized access to an Australian Department of Health site.
- The motion alleges that OpenAI violates the Children’s Online Privacy Protection Act by collecting personal information from children under the age of 13 without parental consent and lacks adequate measures to prevent minors from accessing ChatGPT.
- He adds that “(by) examining how ChatGPT interacts with minors, the researchers “wanted to test (ChatGPT’s) guardrails.” The initial visceral response is, “Oh my Lord, there are no guardrails. » »
Zoom out: The motion cites a report from Axios’ Madison Mills that OpenAI, Anthropic, and security researchers are investigating tens of thousands of incidents involving problematic model behavior, including “circumventing guardrails, creating chat rooms, escaping sandboxes, hijacking websites, self-prompting, or attempting to bypass monitors.”
- The incidents vary in severity and most investigations so far do not appear to have caused any real damage.
- OpenAI recently announced that it was pausing training on its highest-performing models and would only resume when additional safeguards and “alignment improvements” are in place.
Editor’s note: This is a breaking news story. Please check again for updates.
Gn usa
