OpenAI pauses AI model training due to cybersecurity risks

OpenAI has announced a two-week pause in the training of its new artificial intelligence models, citing cybersecurity risks. The company stated that this decision is a precautionary measure to enhance its safety and security protocols.

The pause is reportedly linked to an incident involving OpenAI and Hugging Face, a platform for machine learning models. Additionally, preliminary data suggests that a model under development, codenamed Astra, might achieve critical capabilities in cybersecurity, raising concerns about potential misuse.

OpenAI emphasized that as AI models become more powerful, the risks associated with their development and testing also increase. Therefore, the company believes its monitoring and security standards must evolve to stay ahead of these risks. "We wanted to dedicate the necessary time to achieve these standards, so we temporarily slowed down the pace of scaling," the company said in its statement.

During this two-week hiatus, OpenAI plans to bolster the protection of its research infrastructure, conduct additional testing of its security systems, and expand its monitoring capabilities. For models like Astra and other developments related to cybersecurity, OpenAI is implementing the most stringent protection requirements. These measures include isolating workloads, imposing further network access restrictions, and continuous security system testing.

“

We wanted to dedicate the necessary time to achieve these standards, so we temporarily slowed down the pace of scaling.