AI Security Blog: Prompt Injection, Guardrails & AI Cyber
Research and guides on prompt injection, agent and tool security, guardrails, PII leakage, abliterated and uncensored models, and AI in offensive and defensive cyber.
Prompt injection
- What Is Prompt Injection? The Complete Guide: Prompt injection is an attack where text an LLM reads overrides the instructions it was given. How it works, real incidents, and how to limit the damage.
Agent and tool security
- AI Agent Security: The Complete Guide: AI agent security is protecting LLM agents that read untrusted content and take real actions. The threats, the architecture patterns that work, and a checklist.
Guardrails and defenses
- LLM Guardrails: What They Are and How to Choose Them: LLM guardrails are checks outside the model that inspect what goes in and comes out. The types, where they sit, open and managed options, and how to choose.
Data leakage and PII
- LLM Data Leakage: How Sensitive Data Escapes AI Apps: LLM data leakage is sensitive data escaping through an AI app's prompts, outputs, logs or tools. The seven paths it takes and the controls that close each one.
Abliterated and uncensored models
- What Is an Abliterated Model? Refusal Removal Explained: An abliterated model is an open-weight LLM whose refusal behavior was removed by editing its weights. How abliteration works, what it costs, and why it matters.
AI for cyber offense and defense
- AI in Cybersecurity: How Attackers and Defenders Use LLMs: How attackers and defenders actually use AI in cybersecurity: what threat reports show about misuse, where LLMs help defenders, and what is still hype.
AI vulnerabilities
- OWASP Top 10 for LLM Applications, Explained: The OWASP Top 10 for LLM Applications explained: all ten risks from prompt injection to unbounded consumption, with real examples and the controls for each.