I assess LLM applications for prompt injection, jailbreaks, data leakage, unsafe tool usage, and instruction-conflict vulnerabilities.

This service is designed for teams building chatbots, RAG applications, AI agents, internal copilots, and tool-using assistants.

The audit may include:

Direct and indirect prompt-injection testing

Jailbreak and instruction-conflict testing

System-prompt extraction attempts

Sensitive-data leakage checks

Malicious document and web-content testing

Tool-use and agent behavior review

Input validation and output safety analysis

Risk classification and severity scoring

Mitigation recommendations and retest plan

You will receive a structured security report containing tested scenarios, observed behavior, risk level, reproduction examples, and recommended mitigations.

I can also help implement guardrails, input/output filters, PII protection, safer tool permissions, and automated red-team evaluation datasets.