Built for enterprise security and compliance

Atomicwork has been designed to protect your data at every step along the way.

• Every customer's data is stored in a self-hosted vector database in their own tenant.
• Customers can control what data Atom crawls and indexes in knowledge sources and conversations.
• No user data is used for model training or tuning. We do not share it with third parties either.
• Personally identifiable data (PII) is redacted during data ingestion and masked in requests and Atom conversations.
• AI insights and performance reports are built using anonymized  data to maintain user privacy.

Atomicwork's ensemble AI model architecture is self-hosted on a secure cloud platform and designed to leverage the advantages of different models for different service management and employee experience usecases.

• Responsible AI methods with continuous testing for prompt injections, leakage, model jail break and safety evasion.
• Grounded by customer data and preferences, with boundaries and guardrails to avoid hallucinations and remove biases.

Atomicwork has built layers of security measures to make sure user data is safe, secure and properly used.

• Access control lists (ACLs) are inherited from knowledge management systems or configured in Atomicwork for all content.
• Comprehensive audit logs detail who accessed specific data and all operations performed within the platform.
• Enterprise work applications are accessed only using a self-hosted iPaaS so data stays in the Atomicwork trust boundary.

Atomicwork employs extensive testing techniques to ensure AI models perform reliably under various conditions.

• Continuous monitoring systems to track AI performance in real-time.
• Controlled release cycles so the team can efficiently manage and roll out changes.
• Troubleshooting mechanisms put privacy first and ensure that developers are given only access to redacted data.

‍