The model promises more transparency on the halls of hallucinations and harmful content
Opena Launched a new web page Hub of security evaluation In public sharing information related to the illusion rates of its models. If a model produces harmful content, the hub also highlights how well it behaves as prescribed and tried jail brakes.
This new page provides additional transparency on a company called Opena, for context, faced Multiple lawsuits It is alleged that copyrighted material was illegally used to train its AI models. Oh, yes, and it is mentioned The New York Times Tech makes the company claim Unexpectedly removed the evidence In the newspaper robbery case against this.
The hub of security evaluations is intended to extend on Openay’s system cards. They not only describe the safety measures of development during the launch, but the hub must provide ongoing updates.
“As the science of the AI evaluation develops, we are aiming to share our progress in developing more scalable means to measure model capacity and security,” Openai said in his statement. “Here, it is easy to understand the security performance of the openi systems over time, and that it will support community efforts over time- to increase transparency throughout the field.” Opena adds that it is his work to have more active communication in the area throughout the company.
Introducing Hub of Security Evaluation -A resource to explore security results for our models.
System cards are initially sharing security measures, and the hub is periodically updated as part of our efforts to communicate in advance of security.https://t.co/C8NGMXLC2Y
– Opena (@openai) May 14, 2025
Interested parties can see each sections of each hub and see information on relevant models by GPT-4.1 4.5. The information provided in this hub is “Snapshot” and interested parties should look at its system cards. Messages and other releases for more details.
One of the biggest ones Butts The total security evaluation hub is that Openai is the company that performs these tests and what information to share in public. As a result, there is no way to guarantee that the company will share all its problems or problems with the people.
