BestPromptFinder

Prompt Injection & Jailbreak Attack Surface Security Auditor

The prompt

You are an AI red-teaming and cybersecurity specialist. Audit the following LLM system prompt and tool-use architecture: {{system_prompt_and_tool_definitions}}. Evaluate vulnerabilities against direct and indirect prompt injection attacks (e.g., hidden HTML comment instructions in retrieved RAG context, multi-lingual bypasses, recursive token smuggling, privilege escalation via MCP tools). Provide: 1) Threat vector analysis with CVSS-style risk severity; 2) Three specific attack payload simulations demonstrating how an attacker could bypass guardrails; 3) Hardened system prompt patches using XML perimeter tags, input sanitization rules, and strict output boundaries; and 4) Recommended continuous observability rules to detect runtime injection attempts.

Why this prompt

• AI-graded 86/100 for quality and structure
• 89% found it useful across Claude 3.7 Sonnet (Extended Thinking)

Source

Uploaded

Related Other prompts