Security and Compliance

Appian's AI capabilities are designed to meet enterprise security and compliance requirements. These features operate within Appian's certified cloud infrastructure and adhere to the same robust compliance standards as the core Appian platform, including SOC 2, HIPAA, PCI DSS, and FedRAMP.

Appian uses a defense-in-depth approach to AI security, combining structural platform controls with content-layer protection:

• Structural controls: Initiator-based access enforcement, data fabric security inheritance, and process model execution boundaries.
• Content-layer controls: AI Guardrails evaluate the text content of AI interactions for policy violations, applying environment-wide content safety policies to all generative AI executions automatically.

Certifications and standards

Appian Cloud maintains compliance with industry-leading security standards, including:

• SOC 2 Type II
• ISO 27001
• HIPAA
• PCI DSS (for PCI-compliant sites)
• FedRAMP (for GovCloud environments)

AI capabilities inherit these certifications and operate within the same compliant infrastructure. For the most current information about Appian's compliance certifications, see the Appian Trust Center.

Administrators can also disable a model provider at the environment level to meet supply chain compliance requirements, such as certifying that the environment does not invoke models from a specific provider.

Data handling and privacy

When you use Appian's AI capabilities:

• Your data remains within your designated region unless you explicitly configure cross-region processing using inference profiles.
• Data transmitted to AI services is encrypted in transit and at rest.
• Appian does not use your data to train or improve underlying AI models.
• You maintain full control over your data and can delete it at any time.

For more information about data privacy, see Private AI.

AI intellectual property and usage

Appian is constantly working to improve your experience by developing advanced AI technology. This includes the prompt templates, system instructions, and orchestration logic that power our AI features.

Keep the following guidelines in mind regarding Appian AI technology:

• Ownership and access: All AI configurations, prompt logic, and data structuring methods created by Appian remain our exclusive property. We provide these to you as part of your subscription to help you build and use your applications. Any configurations you create and data you process using Appian AI capabilities remain your property.
• Confidentiality: We consider our AI methodologies to be confidential. You may access them only through your authorized Appian environment.
• Proper usage: You should use Appian AI technology only to support your work within Appian. Do not copy, extract, or use our AI instructions to train, fine-tune, or prompt any AI models outside of Appian.
• Technical integrity: Do not attempt to reverse engineer, disassemble, or determine the source code or development methods of our AI systems.

For more information on how we manage your data within these systems, see our Data Privacy documentation.

High availability

Appian's AI capabilities run on resilient infrastructure that keeps your AI features up and running. We use AWS's multi-availability zone architecture and automatic failover to minimize disruptions and keep things operational, even when infrastructure issues occur.

Data residency and cross-region processing

Your data remains within your designated region unless you explicitly configure cross-region processing using inference profiles. Inference profiles define pre-configured sets of regions where AI models can process data.

When you select an inference profile in the Admin Console, you control which geographic regions can process your AI data. This provides flexibility for data governance while ensuring access to the most powerful models.

For detailed information about inference profiles and regional availability, see Regional Availability.

AI service architecture

This section provides a technical overview of how AI requests are routed and processed. Understanding how data flows between regions helps you make informed decisions about inference profile configuration and compliance requirements.

Our AI architecture is designed with private AI as the foundation and upholds those principles. Learn more about private AI features.

Single region architecture

Appian's AI architecture utilizes a single AWS region to provide a secure and scalable AI as a service solution. These capabilities are in-line with the architecture of Appian Cloud deployments, which also use AWS as the provider.

Scroll or drag to pan · ⌘+/⌘− to zoom · Esc to close

Multi-region architecture

If your site is not located in a region that supports a feature, you can select an inference profile to send your data to a supported region.

Scroll or drag to pan · ⌘+/⌘− to zoom · Esc to close

Foundational principles

Foundational principles of our AI architecture apply to single- and multi-region implementations.

Multi-tenant AI service: Appian's multi-tenant AI service is shared among multiple customers in a given region and has multiple layers of control to restrict data access to respective customer sites. The service has site-level controls to restrict data access and encryption/decryption permissions, whether that distinction is a separate customer site or a separate site within a single customer's environment. This shared service approach allows for economies of scale.

Data transit and retention:

• All of your data stays within the Appian Cloud environment.
• All data in-transit is encrypted using TLS.
• Communication between Appian and the AI service uses asymmetric signing where each customer site has a unique key used to verify its identity.
• Communication between the AI service and Amazon Bedrock uses AWS Signature V4 authentication. When using Appian-managed credentials, the AI service authenticates on your behalf. When using customer-managed credentials, the AI service uses your provided credentials to authenticate to your AWS Bedrock account.
• Communication between the AI service and Amazon Textract uses AWS Signature V4 authentication to verify the identity of the requesting service.
• Bedrock and Textract are stateless, so no information is retained for the majority of models. However, certain models hosted on Amazon Bedrock are subject to AWS's abuse detection policies, which may retain classifier-flagged inputs and outputs for up to 30 days for automated abuse review. This retention is handled entirely by AWS within the Appian Cloud compliance boundary and is not shared with third-party model providers. No data is used to train models.

Learn more about Amazon Bedrock, Amazon Textract, and Amazon Bedrock data retention.

AI data processing and security guarantees

Appian encrypts all AI requests in transit using TLS. When inference profiles route requests across regions, your data stays on AWS's secure private network, it never touches the public internet. Prompts and responses are processed in memory only and are not stored in the destination region.

AI model provider credentials

Appian supports the following credential options for accessing generative AI:

• Appian-managed credentials: By default, Appian provisions and manages the AWS credentials used to access Bedrock foundation models. Appian handles all authentication, credential rotation, and infrastructure management.
• Customer-managed AWS credentials: You can configure your own AWS account credentials in the Admin Console to connect directly to your Bedrock resources. This gives you greater control over security, compliance, and cloud resource management.
• Customer-managed Azure credentials: You can configure your own Microsoft Azure OpenAI account in the Admin Console to route LLM inference through your Azure environment.

When you use customer-managed credentials, the AI service routes requests through your account while maintaining all other security controls. Your credentials are encrypted at rest, transmitted over HTTPS, and never written to server logs.

Credential protection

Appian handles your AWS credentials with multiple layers of security to ensure they are never exposed:

• Secure transmission: All credentials are transmitted over HTTPS.
• Masked storage: The Access Key and Secret Access Key are masked after saving and are never displayed again in the Admin Console.
• No logging: Sensitive credentials are never written to server logs.

Connectivity validation

Before saving your configuration, the TEST CONNECTION feature performs a real-time validation against your AWS account. This confirms:

• Authentication: Your Access ID and Secret Access Key are valid for the selected region.
• Permissions: Your IAM user has the required bedrock:InvokeModel permissions.

Azure OpenAI service limitations

Configuring Microsoft Azure OpenAI as a provider applies only to large language model (LLM) inference for generative AI features. The following services always use Appian-provisioned infrastructure, regardless of which LLM provider is configured:

• Document extraction: Amazon Textract continues to process documents for optical character recognition (OCR) in document extraction AI skills. Your configured LLM provider (such as Azure OpenAI) is used for the LLM inference portion of generative AI skills that take document inputs.
• Knowledge bases and vector search: OpenSearch continues to index data for document-based chat components and smart search.

Data sent to these Appian-provisioned services remains within the Appian Cloud compliance boundary. Data sent to your Azure OpenAI instance for LLM inference is governed by your organization's Azure agreement and data residency configuration.

Shared responsibility model

When using customer-managed credentials, responsibility is shared between Appian and you.

Appian secures:

• The AI service orchestration layer
• Data handling and encryption in transit
• Process execution and monitoring
• Multi-tenant isolation controls
• Appian-provisioned services (Amazon Textract, OpenSearch) regardless of LLM provider

You are responsible for:

• IAM policy configuration and permissions (AWS) or API key management (Azure)
• Credential rotation and lifecycle management
• AWS Bedrock or Azure OpenAI account governance
• Data residency configuration within your Azure environment
• Compliance with your organization's cloud policies

Regardless of which credential model you use, Appian maintains its compliance commitments and does not train models on your data or share your data with third parties.

AI feature security

Each AI feature in Appian has specific security considerations based on how it processes and stores data. This section covers the security details for AI agents, Document Classification and Document Extraction AI Skills, data fabric chat components, document-based chat components, and smart search.

AI agents

AI agents follow all of the foundational AI security principles including the principles of the tools they use. When an AI agent invokes a tool, it checks the initiator's access. If the user lacks permission, the AI agent doesn't run. AI agents always operate with the initiator's access and can't exceed user permissions or bypass object-level security.

Document Classification and Document Extraction AI Skills

The Document Classification and Document Extraction AI Skills use Amazon Textract to recognize the content within the documents.

When a skill is called during training, testing, or process execution, the associated documents are uploaded to an S3 bucket. The documents are kept in the same region as the customer site.

Additionally, the bucket is:

• Encrypted using keys specific to each customer site
• Segmented for each customer site
• Not accessible by application developers or users

All data used for a model both at training and execution is retained until the AI skill is deleted. The model package is deleted when the AI skill object is deleted.

When a user submits a reconciliation task after document extraction, the updated key-value pair mappings are entirely stored within the application on the customer's site.

Learn how document extraction works in Appian.

Data fabric chat components

Appian AI Copilot powers multiple components that allow business users to ask questions of their data fabric, including:

• Records Chatbot
• Data Fabric Chatbot
• AI Copilot for reports
• AI Copilot for data fabric

When using these components, users can only chat with records they have access to. Record-level security is the responsibility of the customer to develop and maintain. No records data is retained in either the AI service or Bedrock due to their stateless nature.

Document-based chat components

Because Enterprise Copilot and Documents Chatbot use retrieval-augmented generation (RAG) to enable users to ask questions about documents, these features index and persist data in Bedrock. Documents used in either of these features are uploaded to an S3 bucket, and processed by Bedrock to create and store vector embeddings in an OpenSearch-based vector index. The documents and vector embeddings are kept in the same region as the customer site. Although documents and embeddings are persisted and stored in S3, chat prompts and responses remain stateless. Prompts or responses are not logged, stored, used to train models, or shared with third parties.

Additionally, the bucket is:

• Encrypted using keys specific to each customer site
• Segmented for each customer site
• Not accessible by application developers or users

Deleting documents

Documents are deleted in S3, along with the related vector embeddings in OpenSearch, in the following situations:

• A document is deleted from Enterprise Copilot
• A user hasn't chatted with a document for 30 days

Smart search

Smart search doesn't require generative AI to answer queries. Instead, it uses embeddings to enable semantic matching. When you enable smart search, Appian uses Bedrock to create vector embeddings from your data source and stores those embeddings in a managed OpenSearch-based vector index. Vector embeddings are kept in the same region as the customer site.

Smart search follows all of the foundational AI security principles. Additional key aspects of smart search architecture and compliance include:

• Indexed fields: Only the fields you configure for smart search are embedded.
• Retention: When you disable fields that you previously selected for smart search on a record type, or delete individual records like text or documents for fields that are currently enabled for smart search, Appian deletes the document in S3 and the related vector embeddings in OpenSearch. Smart search deletions can take up to one week (7 days).
• Access controls: Smart search respects Appian record-level and field-level security; users only see results they are authorized to view.

Feedback