Duration: 3:28

4 views

1

This vid explains OpenAI safety.

i. OpenAI's abuse monitoring system for Azure OpenAI Service is designed to detect and mitigate instances of misuse that may violate the Code of Conduct or other product terms. Here are the key components of the system:

- **Content Classification**: Classifier models identify harmful language and/or images in user inputs and outputs, categorizing them based on severity levels detailed on the Content Filtering page.
- **Abuse Pattern Capture**: The system analyzes customer usage patterns using algorithms and heuristics to spot potential abuse indicators, considering factors like the frequency and severity of harmful content detection.
- **Human Review and Decision**: If content is flagged by the above methods, authorized Microsoft employees may review the content using Secure Access Workstations, with access granted only via Just-In-Time approval by team managers. For resources deployed in the European Economic Area, reviewers are located within the same area.
- **Notification and Action**: Customers are informed via email if a threshold of abusive behavior is confirmed. They are typically given a chance to explain or correct the behavior, with failure to do so potentially leading to suspension or termination of access to Azure OpenAI resources.

Data privacy is a critical aspect, and details on how data is handled can be found on the Data, Privacy and Security page. The system stores all prompts and generated content securely for up to 30 days to facilitate abuse monitoring.

Learn more:
(1) Azure OpenAI Service abuse monitoring - Azure OpenAI | Microsoft Learn. https://learn.microsoft.com/en-us/azure/ai-services/openai/concepts/abuse-monitoring.
(2) Data, privacy, and security for Azure OpenAI Service. https://learn.microsoft.com/en-us/legal/cognitive-services/openai/data-privacy.

Learn more@ https://www.youtube.com/c/ITGuides/search?query=OpenAI.