WASHINGTON, August 20 (WAM) – OpenAI has announced a new set of security and safety policies and procedures aimed at containing security incidents during the testing of artificial intelligence models.

The measures include tightening network isolation practices to prevent a breach in one workload or supporting service from granting unauthorized access to the internet or internal networks, as well as an advanced monitoring system that tracks activities, tools, and logs of the models to detect any concerning behavior or attempts to bypass security restrictions.

OpenAI stated that the new measures were partly in response to the advanced capabilities of the upcoming "Astra" model's cybersecurity features, as well as the continuous acceleration in the development of artificial intelligence technologies.

The company aims to issue alerts within 30 minutes of detecting any concerning activity. It estimates that operating the monitoring system will consume computing resources equivalent to around 20% of those used in the monitored process.

- Khla -