Defense

A practical, defender-first look at how attackers misuse large language models — phishing, malware assistance, reconnaissance and social engineering — and the concrete defenses that hold up: phishing-resistant authentication, detection and user reporting.

Prompt injection is the top security risk for LLM applications: attacker text that hijacks a model's instructions. This explains direct vs indirect injection, why it is hard to fully prevent, and how prompt scanning fits into a layered defense.