Fears of access to sensitive information inside 'OpenAI' and the company defends its measures

A report revealed that employees at 'OpenAI' warned of weaknesses in internal security procedures that could allow access to sensitive technical information, while the company asserts that it has strengthened protection systems and separated its most sensitive operations.

A report revealed that employees at 'OpenAI' warned its management years ago, before the launch of 'ChatGPT', of weaknesses in internal security procedures that could allow access to sensitive technical information. The agency quoted five informed sources that concerns included the breadth of access permissions to information within the company, and the resulting risks related to protecting artificial intelligence research and technologies. In its response, 'OpenAI' confirmed that its sensitive operations are subject to separate protection procedures, and that artificial intelligence model weights are not stored in the unified software repository used by employees. These statements coincide with the company's announcement of new measures to enhance the security of its systems, including hiring external researchers and expanding the use of artificial intelligence models in detecting and addressing security vulnerabilities. The company also announced on Monday preparations for a new round of security tests in cooperation with 'Cathedral Systems', a cybersecurity company, with its researchers working alongside company teams to test systems and identify weaknesses. The company explained that the cooperation aims to develop a continuous mechanism for detecting and fixing vulnerabilities, instead of relying on limited security tests that end once updates are released.