XSS from LLM HTML Output

Trusting Model Output for Authorization

Blocking Sensitive Data Exfiltration

Sandboxing Model-Generated Code

Allow-List Before Downstream Action

Guarding Against Jailbreak Overrides

Masking PII in Model Output

Opting Out of Provider Training

Sanitizing URLs to Prevent SSRF

Excessive Agency and Confused Deputy

OWASP Top 10 for LLM Apps

GDPR & HIPAA for LLM Data

Vendor Data Retention & Opt-Out

Injection Risks in RAG Documents

Redact PII Before Embedding

Least-Privilege Design for Tool Access

Trusted Instructions vs Untrusted Data

Wrong Role for User Data

Over-Privileged Tool Access

Treating Documents as Instructions

Indirect Injection via Retrieved Content

Logging Raw Sensitive User Data

Logging API Keys in Prompts

Ungated Tool Execution

Multi-Tenant Data Isolation

RAG Injection via Retrieved Docs

Provider Data Retention Policies

Encryption In Transit and At Rest

Data Exfiltration via Outputs

Trusted vs Untrusted Input Separation

Delimiting Untrusted Content

System Prompt Leakage

Direct vs Indirect Prompt Injection

Excessive Agency in Agents

Least-Privilege for LLM Resources

Preventing Secret Leakage in Prompts

Data Exfiltration via LLM Outputs

System Prompt Leakage and Extraction

Direct vs Indirect Prompt Injection