Openai promises more transparency in model hallucinations and harmful content
Openai has launched a new webpage called the Security Assessment Center to publicly share information such as model hallucination rates. The hub will also highlight whether the model produces harmful content, how it is represented and how it is jailbreaking.
The tech company claims this new page will provide additional transparency to OpenAI, a company that faces multiple lawsuits accusing it of illegally using copyrighted materials to train its AI models. Oh, yes, worth mentioning The New York Times Claims the tech company accidentally deleted evidence in a newspaper theft case.
The Security Assessment Center aims to expand OpenAI's system cards. They only outline security measures at development, and the hub should provide continuous updates.
“As the science of AI evaluation evolves, we aim to share our advancements in developing more scalable methods to measure model capabilities and security,” Openai said in the announcement. “By sharing a portion of our security assessment results here, we hope that this will not only make it easier to understand the security performance of OpenAI systems over time, but also support community efforts to increase transparency across the field.” Openai added that its efforts communicate more actively in this area of the company.
Introducing Security Assessment Center – a resource that explores the security outcomes of our models.
While the system card shares security metrics at startup, the hub will be updated regularly as part of our active efforts to communicate about Safety.https://t.co/c8ngmxlc2y.
– Openai (@openai) May 14, 2025
Interested parties can view each part of the hub and view information about relevant models, such as GPT-4.1 to 4.5. OpenAI notes that the information provided in this hub is just a “snapshot” and interested parties should view their system cards. Evaluate and other versions for more details.
big but For the entire Security Assessment Center, OpenAI is the entity that performs these tests and chooses to publicly share information. As a result, there is no way to guarantee that the company will share all issues or concerns with the public.
