AI Data Security
Protecting sensitive data when using AI systems, where every interaction including prompts, uploaded files, tool call results, and agent memory is a potential data exposure point.
Also known as: AI Security, AI Information Security
Category: AI
Tags: ai, safety, data-management, governance
Explanation
AI data security addresses the protection of sensitive information across every touchpoint with AI systems. Every interaction with an AI tool is a potential data exposure point: prompts, uploaded files, tool call results, and agent memory all contain information that may be confidential, proprietary, or regulated.
## Attack surface
The data exposure surface in AI systems is broader than many realize:
- **Prompts**: everything typed or pasted into an AI tool is data being shared with the provider
- **File uploads**: documents, code, and images uploaded for analysis
- **Tool call results**: when AI reads files, queries databases, or calls APIs
- **Agent memory**: accumulated context that persists across sessions
- **Skill content**: instructions that may reference internal architecture, processes, or credentials
- **Output logging**: provider-side logging of conversations and outputs
## Threat categories
### Data leakage
Sensitive data included in prompts reaches the provider's servers. Your data may be used to train future models. Agent memory accumulates confidential information over time. Cross-session contamination can cause context from one project to leak into another.
### Data exfiltration
Malicious skills or prompts can extract data through tool calls. Prompt injection attacks can redirect AI to send data to external endpoints. Untrusted skills accessing sensitive data represent a supply chain security risk.
### Compliance violations
Processing personal data (GDPR, CCPA) through non-compliant AI services, storing regulated data (HIPAA, SOX) in AI memory systems, and cross-border data transfer through cloud AI providers all pose regulatory risks.
## Mitigations
- Classify data before sharing with AI, following clear usage policies
- Use API access with enterprise data handling agreements
- Run sensitive workloads on self-hosted models
- Implement AI agent permissions to restrict file and tool access
- Maintain context isolation between projects and clients
- Audit agent memory for accumulated sensitive data
- Strip credentials and secrets from context before loading
Related Concepts
← Back to all concepts