ONEiO and AI (ONEAi®)

ONEiO is continuously enhancing its product features by incorporating the latest AI technologies, collectively known as ONEAi®. On this page, we will delve into the technical and security aspects of ONEAi®. To stay updated with all our latest developments, please select the "follow" option at the top right corner.

What specific AI technologies and algorithms are being used?

ONEiO utilizes AI in multiple features of the ONEiO service and the selected technologies vary. Common for all of these is that we use AI services hosted in AWS.

Text translation: For translating text, ONEiO uses Amazon Translate. Amazon Comprehend is used to automatically detect the source language when it’s not configured.

Mapping suggestions: ONEiO uses foundation models for generative AI through Amazon Bedrock service. The specific AI models can vary, but currently we use Anthropic Claude 3 models. Retrieval Augmented Generation (RAG) is being used to give the AI model access to our knowledge base that has information about existing mappings in ONEiO.

How is data security maintained when using AI?

ONEiO uses only AI models that are hosted on AWS, which is where ONEiO service itself also resides. For our EU location, AI models are hosted within the EU and for the US location within the US. Any customer content processed by the models is encrypted and stored at rest in the AWS region that ONEiO has selected. The inputs and model outputs are not shared with any model providers and are accessible only by ONEiO and AWS.

What security and compliance do the AI solutions support?

ONEiO AI solutions are in the scope of ONEiO’s ISO 27001 certification. The underlying solutions from AWS are in scope for common compliance standards such as SOC, ISO, HIPAA and GDPR.

Does ONEiO or its sub-processors use customer data to train/improve the AI models?

The AI models that ONEiO utilizes are not trained by customer data by us or by any of our sub-processors. For mapping suggestions, we do use a hand-curated list of existing integration field mappings, but we make sure that no data from the integration messages themselves is used.

What are the privacy implications of using AI?

For mapping suggestions, we use a hand-curated list of existing integration field mappings. The hand-curated list does not contain any PII data.

For text translation, the customer is responsible for evaluating whether passing PII data in the integration message fields that are to be translated is ok. The customer also is in control of what fields, if any, are to be translated.