Stop Indirect Prompt Injection

Use Case

Indirect prompt injection hides malicious instructions inside content that an AI system later retrieves and trusts. PointGuard AI inspects that content before it reaches the model, closing a critical gap in RAG and agentic workflows.

Challenges

The AI environment creates several recurring security and governance challenges:

• Malicious instructions hide inside trusted retrieved content
• Users may never see the injected prompt
• Documents, code, PDFs, web pages, and metadata create exposure
• Injected reasoning can trigger unauthorized downstream tool calls

Solution

PointGuard AI provides a comprehensive solution for inspecting retrieved content before model ingestion and governing resulting tool actions:

1. Inspect retrieved files. Use AI Intelligent Guardrails and MCP Security Gateway to scan documents, source code, PDFs, and other retrieved content for hidden malicious instructions before they reach the model.

2. Analyze metadata and context. Evaluate metadata and surrounding content for prompt-injection indicators that may not appear in the visible user request.

3. Block compromised content. Prevent malicious retrieved instructions from entering model context or influencing downstream responses.

4. Control resulting tool actions. Apply gateway policies to stop unauthorized tool execution even if an injected instruction influences agent reasoning.

This creates protection on both sides of the model: malicious content is filtered before ingestion and dangerous actions remain governed after inference.

Risks Addressed

Applicable framework risks and controls include:

OWASP Top 10 for LLMs

• LLM01:2026 Prompt Injection
• LLM02:2026 Sensitive Information Disclosure
• LLM08:2026 Hidden Context Exposure
• LLM10:2026 Improper Output Handling

OWASP Top 10 for Agentic Applications

• ASI01: Agent Goal Hijack
• ASI02: Tool Misuse and Exploitation
• ASI05: Unexpected Code Execution (RCE)
• ASI06: Memory & Context Poisoning
• ASI10: Rogue Agents

NIST AI Risk Management Framework

• GOVERN 1.4: Transparent risk policies, procedures, and controls are established
• MAP 1.1: Intended use, context, users, and lifecycle risks are documented
• MEASURE 2.7: AI system security and resilience are evaluated and documented
• MANAGE 4.1: Post-deployment monitoring and change management are implemented