Privacy & Security

Five-fold security commitment

International standard certification

• 100% encryption of all data: All data stored in AtomDeep AI is encrypted using the latest AES-256 algorithm, ensuring that even if the database is compromised, the data remains secure. The encryption key is stored in a secure location and is never transmitted to the client.

No data is stored in cloud memory

• API calls with "read-and-burn" mechanism: When the system needs to call cloud-based LLMs for analysis, data is only held in memory for millisecond-level transient computation. Once the inference task is completed, the data is immediately released and never written to the hard drive of the LLM provider.

Enterprise data is not used for model training

• Data is never used for model training: We clearly distinguish between "inference" and "training". AtomDeep AI only uses your data as temporary reference context for RAG (retrieval-augmented generation), not as training corpus. We are willing to contractually commit that your core formulations and processes will never be used to optimize general-purpose LLMs, eliminating the risk of your core intellectual property becoming fodder for public model training.

Data never leaves the corporate intranet

• Data never leaves the corporate intranet: For customers with extremely high confidentiality requirements, we support a fully on-premise deployment solution. Deploy the LLM and vector database entirely on your enterprise's private servers, physically blocking external network connections, ensuring at the physical level that data never leaves your premises.

Dual protection: legal and permission-based

• Contractual and permission safeguards: Beyond technology, there is legal protection. We sign a strict non-disclosure agreement (NDA) with your company at the start of service. Meanwhile, the software inherits a fine-grained permission system, ensuring that even within the enterprise, sensitive data is only visible to authorized personnel—keeping AI trustworthy with your secrets.