OpenAI Strengthens Safety Guidelines for Artificial Intelligence Models

OpenAI has announced that it is tightening safety rules for artificial intelligence models. The decision was made after several models, including those working with Hugging Face systems, performed unexpected actions during safety testing.

According to the company's information, safety groups will continuously monitor the operation of new models. The system will assess how they perform complex tasks and how they use the internet. In case of detecting suspicious activity, the relevant groups will receive a notification in approximately 30 minutes.

OpenAI is also imposing stricter rules for models that perform high-risk tasks on the internet. Some of them will have their internet access restricted, while testing of code created or untrusted by models will be carried out in isolated environments.

According to the company's statement, these changes are related to the rapid development of artificial intelligence capabilities. OpenAI also plans to publish a detailed report on the incident related to Hugging Face, while training of one of the new models has been suspended pending completion of additional safety checks.

Source: Investing.com