OpenAI just rolled out a new AI safety program, bringing in outside experts from four organizations to review its internal models and processes.
The goal? More transparency around AI risks, especially after the recent Hugging Face controversy.
But not everyone is convinced it'll work, since these evaluators don't get full access or open communication with OpenAI staff.
Sydney Von Arx criticizes OpenAI access
Some researchers, like Sydney Von Arx, founder of AI safety research group Nightingale, are not thrilled about the restrictions.
She points out that assessors cannot freely chat with all OpenAI employees or use platforms like Slack, which clashes with earlier promises of "employee-like access."
OpenAI admits there is no set standard for these reviews yet, but says it plans to improve things going forward.
CEO Sam Altman also stressed at a recent event that third-party checks are key for making AI safer and more transparent.