9.1
9.1 AI Application Security & Adversarial Testing
Direct/indirect prompt injection defense, instruction hierarchy, jailbreak testing, NeMo Guardrails, tool privilege escalation prevention, and safe deserialization.
Prompt Injection Defense, Jailbreak Interception & NeMo Guardrails
Strict instruction hierarchy control (System > Tool > User), perplexity anomaly filters, delimiter tag hardening, NVIDIA NeMo Guardrails (Colang 2.0), and OWASP Top 10 for LLM v2.0 vulnerability mitigations.
🏢 Companies
Lakera、NVIDIA (NeMo Guardrails)、Cisco (Robust Intelligence)、Promptfoo、Protect AI
🛠️ Tech Stack
Prompt InjectionJailbreak DefenseInstruction HierarchyNeMo GuardrailsColangOWASP Top 10 for LLMLakera
💼 Roles & Salary
AI Security Engineer、AI Red Team Engineer、Application Security Engineer
💰 $210K - $450K / year (AI Security & Prompt Defense) | ¥600K - ¥1.4M / year
📚 Prerequisites: OWASP Top 10 for LLM Vulnerability Defense • NeMo Guardrails & Colang 2.0 Policy Authoring • Perplexity Anomaly Detection & Adversarial Attacks • Structured Tag Escaping & Instruction Hierarchy
Tool Privilege Scoping, DLP Egress & Supply Chain Security
Dynamic least-privilege tool token scoping, streaming PII redaction (Presidio), sandbox egress network policies, and model weight malicious deserialization defenses (Safetensors / ModelScan).
🏢 Companies
Protect AI、HiddenLayer、Cisco (Robust Intelligence)、Palo Alto Networks
🛠️ Tech Stack
Tool AbuseData Exfiltration (DLP)PII RedactionEgress FilteringSafetensorsModelScanSupply Chain Security
💼 Roles & Salary
AI Security Engineer、AI Systems Engineer、Threat Researcher
💰 $205K - $440K / year (Tool Scoping & AI Supply Chain Security) | ¥580K - ¥1.35M / year
📚 Prerequisites: OAuth 2.0 Token Scoping & Least Privilege • Streaming PII Redaction & Masking Algorithms • K8s Egress NetworkPolicy & Firewall Filtering • Safetensors Signatures vs Pickle Deserialization Exploits