AI Prompt Injection Pentest
Budget: ₹1,500 – ₹12,500 INR
I’ve just released a production-grade chatbot and now need a seasoned red-teamer to see how far its guard-rails can be pushed. Your job is to run an end-to-end security assessment that zeroes in on prompt injection, jailbreak techniques, and any threat-detection angles relevant to LLM-powered systems.
You’ll have direct access to the public interface (plus limited system-prompt details if that helps). Think like an adversary, chain prompts, exploit context windows, and surface every path that lets a user override hidden instructions, leak data, or trigger unsafe actions.
When you reply, point me to past work that proves you’ve broken—or secured—similar chatbots. Solid examples of exploits or red-team reports are the single most important factor in my selection.
Deliverables
• Methodology or test plan outlining the attacks attempted
• Step-by-step transcripts (and screenshots) of every successful jailbreak or prompt-injection exploit
• Impact and severity rating for each finding
• Clear, actionable fixes and guardrail improvements
• Concise final report (PDF or Markdown)
I’ll sign off once I can reproduce the issues from your steps and confirm that your mitigations close each gap.
You’ll have direct access to the public interface (plus limited system-prompt details if that helps). Think like an adversary, chain prompts, exploit context windows, and surface every path that lets a user override hidden instructions, leak data, or trigger unsafe actions.
When you reply, point me to past work that proves you’ve broken—or secured—similar chatbots. Solid examples of exploits or red-team reports are the single most important factor in my selection.
Deliverables
• Methodology or test plan outlining the attacks attempted
• Step-by-step transcripts (and screenshots) of every successful jailbreak or prompt-injection exploit
• Impact and severity rating for each finding
• Clear, actionable fixes and guardrail improvements
• Concise final report (PDF or Markdown)
I’ll sign off once I can reproduce the issues from your steps and confirm that your mitigations close each gap.