All threats

prompt-injection

Attacker manipulates AI model inputs to bypass safety controls, extract sensitive information, or cause unintended behavior

MITRE ATT&CK techniques

Common Weakness Enumeration

• CWE-1427

Mitigating controls

ctrl-prompt-1

Implement input validation and sanitization for all prompts

ctrl-prompt-2

Use system prompts that establish clear behavioral boundaries

ctrl-prompt-3

Implement output filtering to prevent sensitive data leakage

ctrl-prompt-4

Monitor and log AI interactions for anomalous patterns

ctrl-prompt-5

Apply rate limiting per user and session

ctrl-prompt-6

Implement platform-specific content safety guardrails (e.g., Bedrock Guardrails, Azure Content Safety, Vertex AI Safety Settings)

ctrl-prompt-7

Isolate user-provided input from system instructions using structured API formats

References

• https://genai.owasp.org/llm-top-10/
• https://atlas.mitre.org/