Security Measure from OpenAI in AI Model Development Process
Following the new model Astra approaching cybersecurity capability thresholds and recent incidents, OpenAI has decided to slow down the pace of model development.
As a result of the OpenAI-Hugging Face incident and preliminary findings that the new model named Astra could reach cybersecurity capability thresholds, the company has slowed down model development and increased security measures.
New Security Measures in the Development Process
Incidents in recent weeks and approaching critical thresholds have brought a sense of urgency to strengthening monitoring, compliance, and containment measures in artificial intelligence systems.
Temporary Pause in Scaling Activities
OpenAI has temporarily slowed down its scaling pace, including a two-week pause on reinforcement learning training conducted on models targeted for latest deployment.
Securing Research Environments
Practices such as workload isolation, network isolation, and continuous security testing have been implemented, mandating the strictest security measures for Astra and cyber-connected workloads.
Advancing Monitoring and Compliance Research
A multi-stage monitoring architecture was established using activation classifiers and automated researchers, ensuring alerts are issued for suspicious activities within thirty minutes.