AI in Qlik Cloud
Design Principles for AI on Qlik Cloud
Qlik believes the productivity and efficiency gains AI provide should not require customers to take risks with their data. As a company, Qlik's approach to AI is based on these key principles:
- Reliability
- Customer Control
- Transparency
- Explainability
- Inclusive
These principles are explained at our Trust & AI page. The design principles drive our development efforts and ensure the solutions we deliver meet Qlik’s ethical standards, giving our customers confidence that leveraging AI with Qlik does not compromise their data security & governance needs.
Based on Qlik's overarching principles, AI principles and governance requirements, the following design principles apply to our AI initiatives at Qlik:
- Qlik does not use the customer’s data to train any AI model external to the customer’s tenant. Any training of models internal to the customer’s tenant, such as using the machine learning aspects of Qlik Predict, is exclusively to the benefit of that customer and is not shared with others.
- Qlik Cloud is a highly governed and audited platform, holding many certifications and accreditations (see Qlik Trust & Security). Qlik complies with the relevant laws applicable including the EU AI act for Limited risk and General risk use-cases. Any design decisions must align with the requirements of these laws, certifications and accreditations.
- AI compliance reviews are conducted as part of our product development processes to ensure our AI implementations align with our general AI principles.
We also apply our general cloud design principles to AI, namely:
- Qlik Cloud is a no view service. This means neither Qlik employees, our cloud providers, nor any parties other than the customer themselves can see the customers data. Our AI initiatives must meet this requirement also.
- Qlik cloud uses multiple layers of encryption (optionally using customer provided encryption keys) for all data stored on the platform including AI & unstructured data.
- All data moving into and out of Qlik Cloud (including AI) is secured with TLS.
Guardrails
When interacting with large language models (LLMs) Qlik applies a number of guardrails to minimize risks for our customers:
- LLM monitoring for harmful content. Qlik scans and rejects questions that contain invisible text or prompt injections.
- Hallucinations. Qlik mitigates hallucinations by detecting contextual relevance in the provided content. Additionally, users can view the content used as a source to verify responses.
Data Sovereignty
When a Customer creates their Qlik Cloud tenant, they are able to choose the region their tenant resides in. A customer's data does not leave their chosen region by default. Due to the speed of rollout of some AI services, as well as resource constraints from our our providers, certain features may necessitate limited data to be sent securely to another region. Where this is required, customers must opt-in to this. Qlik will not transfer data to other regions without the customers express authorization.
Cross-region inference
Certain AI features in Qlik cloud require customers to allow Cross-region data processing in a very limited capacity in the use of large language models (LLMs). This is known as cross-region inference, or CRIS for short. Enabling cross-region inference allows Qlik Cloud to temporarily send the data required for processing a question to the AWS region where the service is available, process the request, and return the results. Enabling cross-region inference does not give Qlik or AWS access to your data. This requirement exists due to the high demand for AI services and limited availability of hardware from AWS. This is not unique to Qlik and is required for most AI services today due to industry-wide resource constraints.
What does this mean for my data?
- All data at rest including tenant event logging information is stored encrypted in the Customer's tenant in the customers chosen region using the customers own encryption keys.
- Data sent between AWS regions is encrypted with AWS's Encryption keys and travels through the secure AWS private network with end-to-end encryption for data in transit.
- No data is persisted in the other regions.
More information on how this impacts specific AI services is covered in the relevant sections of this document. For a detailed underatanding of these processes at AWS, see Securing Amazon Bedrock cross-Region inference
Authorization & AI in Qlik Cloud
AI services in Qlik cloud fully respect any configured authorzation in Qlik Cloud. Access to applications, files and connections rely on permissions granted to the spaces those assets reside in. Within Applications, our AI services are bound by any section access configured for that application. If a user is restricted from data when using the application in a conventional way, they will also be restricted when interacting though AI.
AI Services in Qlik Cloud
AI is used throughout Qlik Cloud in many areas. Broadly speaking, our AI services fit into two key areas:
- Specific AI services and capabilities in Qlik Cloud
- AI Accelerators and helpers for non-AI services and capabilities in Qlik Cloud AWS change and replace services over time, so it is possible Cross-region inference requirements may change in the future. The following is true at the time of writing.
The following AI services in Qlik Cloud do not require Cross-region inference to function:
- Qlik Answers (legacy assistants only)
- Qlik Predict
The following AI services are available in Qlik Cloud that rely on Cross-region inference to function:
- Qlik Answers (excluding legacy assistants)
- MCP Server
- Discovery Agent
- Qlik Talend Cloud Pipelines: AI SQL generation and AI model suggestions (see Generating a SQL transformation from a text prompt).
- Data Products: AI description generation (see Working with validation rules .) and AI DQ Validation rule suggestions (see Generating an AI-based description) .
AI Models at Qlik
Qlik selects AI models and CRIS inference profiles to best fit with our customers Data Sovereignty, data security and geo-political requirements wherever possible. This is bounded by the limitations enforced on us by AWS - i.e. what infrastructure and CRIS profiles are made available for each Qlik home region by AWS. This is under constant review by AWS and Qlik to best serve the needs of our customers. For the latest breakdown of potential inference processing regions by product feature and home region please see: Enabling cross-region data processing - Qlik Help.. You can also subscribe to the following community page to receive updates to processing regions when changes are announced: Qlik Cross-Region Data Processing - Qlik Community .