# AI Data Security
Protecting sensitive data when using 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
- **Prompts**: everything you type or paste into an AI tool is data you're sharing
- **File uploads**: documents, code, images uploaded for analysis
- **Tool call results**: when AI reads your files, queries your databases, or calls your 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
- [[AI Training Data Collection]]: your data may train future models
- Agent memory accumulates confidential information over time
- Cross-session contamination: context from one project leaking into another
### Data exfiltration
- Malicious skills or prompts that extract data through tool calls
- [[Prompt injection]] that redirects AI to send data to external endpoints
- [[AI Skill Supply Chain Security]] failures: untrusted skills accessing sensitive data
### Compliance violations
- Processing personal data (GDPR, CCPA) through non-compliant AI services
- Storing regulated data (HIPAA, SOX) in AI memory systems
- Cross-border data transfer through cloud AI providers
## Mitigations
- Classify data before sharing with AI (see [[AI Usage Policy]])
- Use API access with enterprise data handling agreements
- Run sensitive workloads on self-hosted models ([[Running AI Models Locally]])
- Implement [[AI Agent Permissions]] to restrict file and tool access
- [[Context Isolation]]: separate contexts between projects/clients
- Audit agent memory for accumulated sensitive data
- Strip credentials and secrets from context before loading
## References
-
## Related
- [[AI Privacy]]
- [[AI Training Data Collection]]
- [[AI Usage Policy]]
- [[AI Agent Permissions]]
- [[AI Governance]]
- [[AI Safety]]
- [[Context Isolation]]
- [[Context Poisoning]]
- [[AI Skill Supply Chain Security]]
- [[Prompt injection]]
- [[Running AI Models Locally]]
- [[Enterprise Context Management (ECM)]]