OpenAI fired three employees who, according to the company, violated rules on handling confidential information. Anecdotal evidence suggests that these employees shared sensitive information with third-party organizations that assess the safety of artificial intelligence.
Source image: Mariia Shalabaieva / unsplash.com
“Our investigation established that these employees mishandled confidential information by circumventing established company procedures. In doing so, they violated our internal rules and undermined the trust required to do our job.”– an OpenAI representative said in a CBS News commentary.
The company added that the audit found systemic violations by employees who had access to confidential information and used the company’s research. A company representative pointed out that OpenAI experts responsible for security have information that requires a high level of trust, without which internal interactions are impossible. The company did not specify what information the breaches related to or what they were.
Three employees — two researchers focused on artificial intelligence safety and consistency, and a research project manager — reportedly shared confidential company information with outside groups evaluating artificial intelligence systems. That’s according to a person familiar with the matter who spoke on condition of anonymity because we were talking about internal company matters. He said some of the allegedly mishandled information involved OpenAI’s infrastructure.
OpenAI has come under scrutiny after its artificial intelligence agents escaped testing environments and compromised the Hugging Face portal, an artificial intelligence model repository. Earlier this month, the company also reported six cases of “unexpected or concerning” system behavior.
In one case, a yet-to-be-released research model added instructions to its work notes to bypass restrictions, while in another case, an artificial intelligence agent uploaded a file online without the user’s knowledge to retrieve links from the browser, according to OpenAI. The company said the six incidents were discovered during model training or testing in recent months.
In addition, OpenAI’s artificial intelligence model also accesses public information from the U.S. Securities and Exchange Commission (SEC) website and data from the U.S. Census Bureau. OpenAI said it found no signs of a data leak or vulnerability.
OpenAI previously decided to cancel the release of new AI models due to security concerns. Saachi Jain, head of the company’s security systems business, said that GPT-6.1 Astra “doesn’t hit the mark in terms of compliance, authoritativeness and how it communicates to users what operations are being performed.” The U.S. Federal Trade Commission (FTC) said on Wednesday it has launched an investigation into OpenAI, Anthropic and other artificial intelligence companies into the risks their technologies may pose to consumers.
If you find an error, select it with your mouse and press CTRL+ENTER.
