AI Security and Data Privacy at Wazoku

Core Security and Privacy Measures

1. 1. No Public Model Training:

   Wazoku does not use client prompts or data to train public language models. Any interaction with the LLM is contained entirely within the scope of the project, meaning there is no feedback loop that would send data back to the broader training set.

2. 2. Data Isolation:

   When using Microsoft Azure’s OpenAI services, data remains isolated. Azure provides built-in encryption at rest and in transit, ensuring that no third parties can access the data during processing. The data submitted by clients is used solely for generating responses, and it is not stored or used to improve general models.

3. 3. Controlled and Private Deployment:

   We leverage Azure’s private deployment options, which allow LLMs to operate within a controlled environment where customer data cannot be accessed or shared externally. This guarantees complete segregation between client-specific instances and other environments.

4. 4. Internal AI Agents:

   At Wazoku, we use a trained AI agent named Jen. Jen is trained on publicly available models and kept entirely internal. This ensures that data processed by Jen is not shared between sites, with external systems or models, and no new data goes back into any public LLM.

5. 5. Azure’s Commitment to Data Privacy:

   Azure OpenAI ensures compliance with strict privacy standards, including GDPR and other global privacy regulations. Azure guarantees that customer data is not used for training OpenAI models, a key differentiator when compared to other LLMs available in the market.

6. 6. Explicit Contractual Commitments:

   We provide contractual assurances regarding data handling, emphasizing that all data stays private and is only used for its intended purpose. This legal commitment serves as a strong foundation for data privacy.

7. 7. Rigorous Access Controls:

   Each client interaction is governed by rigorous access control policies, which limit who within Wazoku and Azure can access or interact with data submitted to the models. Additionally, any sensitive data can be encrypted or anonymized before being processed, providing extra assurance that data cannot be identified or traced.

8. 8. Audit and Monitoring Capabilities:

   Wazoku maintains detailed audit logs and monitoring capabilities. These tools offer full transparency, and clients can request insights regarding data processing activities.

9. 9. Commitment to Evolving Security Standards:

   As a company that remains LLM agnostic, we are committed to adapting our solutions as the market evolves. We are open to integrating new LMs or partnering with different AI service providers based on client needs and emerging best practices.