We're excited to introduce AI Guardrails, a new DLP feature that leverages AI to secure AI.
AI Guardrails is an integral part of AI Access, our AI-focused feature set that provides comprehensive visibility and control over the usage of third-party AI tools (often referred to as shadowAI) and ensures these tools are used safely and securely.
How do AI Guardrails enhance AI security and control?
As enterprises increasingly adopt AI tools, managing and securing their use becomes crucial. AI Guardrails address this by providing robust security measures that protect against malicious prompts, unsafe or unethical usage, and ensure compliance with regulations, especially concerning Personally Identifiable Information (PII).
Three AI Guardrails available:
- Security Guardrail – Protect AI applications from threats and unauthorized access, and to prevent their misuse for malicious activities. For example, it can block prompt injection attacks that attempt to bypass AI protections for unauthorized data retrieval or to compromise the AI tool.
- Safety Guardrail – Prevents unethical or unsafe use of AIapplications. For example, it blocks prompts that could lead to self-harm or harm to others, whether physically or mentally.
- Privacy Guardrail – Prevent users from sharing regulated and sensitive data with AI applications.
How can you enable AI Guardrails?
To utilize AI Guardrails, start by leveraging a new DLP rule type called AI Guardrails. Within this DLP rule, select the specific AI Guardrails to apply by choosing from a new type of data classification known as AI Guardrail Classification. There are three built-in AI Guardrail classifications available: Security, Safety, and Privacy. You have the flexibility to use these predefined classifications or customize them to granularly decide what specific elements to monitor for each AI Guardrail. This customization allows for tailored protection that aligns with your unique security and compliance needs.
Image 2 - New AI Guardrails Classifications