OpenAI pledges to publish AI safety test results more often

image via TechCrunch

May 14, 2025, 4:38 PM

• •OpenAI is launching a Safety Evaluations Hub to regularly publish AI model safety evaluation results.
• •The hub will share metrics on an ongoing basis and be updated with major model updates.
• •OpenAI aims to increase transparency and support community efforts in AI safety.
• •The move follows criticism over the company's safety testing practices and model behavior issues.

OpenAI is increasing transparency by launching a Safety Evaluations Hub to regularly publish safety evaluation results for its AI models. The hub will be updated periodically with safety metrics, aiming to improve understanding of OpenAI's systems and support community efforts for increased transparency. This move comes after criticism regarding safety testing of the models and issues with the default model powering ChatGPT, leading to the implementation of new fixes and changes. The company's CEO, Sam Altman, has also faced scrutiny regarding model safety reviews.

Read original article