Frequently Asked Questions¶

Yes, PAIG can be used for any AI application. PAIG is a generic AI governance platform that can be used for any AI application.

Absolutely, PAIG can be configured to operate in an 'observation-only' mode. In this mode, PAIG will monitor and log data access activities without actively enforcing any restrictive policies like Deny or Redact. This approach is particularly beneficial for organizations seeking to first analyze and understand their data access patterns. It allows you to gain valuable insights into user behavior and data usage, which can inform future policy implementations. Additionally, this mode is highly effective for compliance reporting, as it provides a clear picture of data interactions within your environment without impacting normal operations.

PAIG integrates seamlessly with AI applications through two primary methods:

1. Transparent Integration with LangChain: For applications developed using LangChain, integration is streamlined via the Privacera Shield library. This method ensures automated policy enforcement throughout the application, including advanced features like Chains and Retrieval Augmented Generation (RAG).

2. Python Library for Native AI Applications: For AI applications natively built in Python, PAIG offers a dedicated library. This allows direct integration, providing developers with the flexibility to utilize PAIG APIs according to their unique business and technical needs.

For more detailed information on these integration methods, please refer to our Integration section.

PAIG uses a series of detectors. We use AI models to detect PII, Toxic content and Source Code in unstructured content. We will be introducing Prompt Injection and PHI detection in February. Password and Credentials detection in content is also in the roadmap.

Our system is designed to identify sensitive data with a high level of accuracy, leveraging advanced algorithms and continuous data analysis. We are committed to ongoing improvements in our detection capabilities, regularly updating our algorithms and methodologies to enhance precision and adapt to emerging data protection challenges. information.

Our system is optimized for efficiency, ensuring that the process of identifying and masking sensitive data incurs minimal performance overhead. While the exact time can vary depending on content size, we've fine-tuned our approach to keep the delay to a minimum. For typical user queries and within the token limits of Large Language Models (LLMs), the additional time required for masking is generally negligible. Most queries are processed with a very modest increase in response time, and even for larger content sizes, the impact remains relatively low. We continuously work on optimizing our algorithms to balance robust data protection with the need for swift data processing, ensuring our solution scales effectively with your requirements.

Yes, PAIG supports streaming protocols by actively screening data as it streams in. Based on established policies, PAIG performs any necessary redactions in real-time during streaming. If a denial policy is triggered, the streaming will be abruptly terminated. It is essential for GenAI applications to gracefully handle such interruptions by providing appropriate error messages or fallback responses.

PAIG doesn't support synchronization of users from AD/LDAP. Currently users and groups can be only added manually. The option to add users/groups using API in the roadmap.

PAIG doesn't support synchronization of users from SCIM. Currently users and groups can be only added manually. The option to add users/groups using SCIM in the roadmap.

PAIG doesn't support managing users and group using public APIs. Currently users and groups can be only added manually. The option to add users/groups using public APIs in the roadmap.