New here?
Start here.
This desk covers offensive AI security: how prompt injection, jailbreaks and tool-use abuse actually work, what the defenses do under pressure, and which of it survives contact with a production system. This page is the short route in.
Read these first
Six pieces that stand on their own. Start with whichever matches the problem in front of you.
- Many-Shot Jailbreaking: How Long Context Breaks Alignment
- Insecure Output Handling Examples: XSS, SQLi, RCE
- Prompt Injection Attack Compendium (2026 Edition)
- LLM Security Vulnerabilities: What Actually Gets Exploited
- Prompt Injection Examples: A Practitioner's Attack Library
- LLM Jailbreak Techniques: Attack Taxonomy and Defenses
Keep up
One email, Friday afternoon UTC. What landed, why it matters, what to do about it. Unsubscribe in one click.
- Weekly digest, one email
- No tracking pixels, no surveillance, no data sale
For practitioners
- GlossaryTerms of art across offensive, defensive, and ML-ops, defined the way practitioners use them.
- LLM Red Team Engagement ChecklistScoping to reporting in one printable pass: pre-engagement, execution, and reporting gates.
- Prompt Injection Testing ChecklistA repeatable per-target pass for direct, indirect, and tool-mediated injection.
- AI incidents snapshot50 incidents from the AI Incident Database and MITRE ATLAS: filterable, dated, sourced.
- AI Security Tools Directory ↗40+ scanners, guardrails, and monitors compared in one maintained table.
- Browse by topicEvery subject this desk covers, from injection classes to eval reliability.