Red-Teaming

The Local Brain: Security Risks of Self-Hosted AI As we move toward running powerful models like Llama 3 or Gemini local-first in our homelabs, we’re introducing a new class of vulnerabilities. The most dangerous among them? Prompt Injection. What is Prompt Injection? At its core, prompt injection is a “jailbreak” where an attacker provides input that overrides the model’s original system instructions. In a local lab environment, this usually takes the form of Indirect Prompt Injection. ...

Red Teaming in the Age of Autonomy We’ve all seen what LLMs can do for productivity. But in the hands of a skilled offensive security professional, these models become a force multiplier that we haven’t seen since the release of Metasploit. The Good, The Bad, and The Automated 1. Phishing at Scale The days of “broken English” phishing emails are over. LLMs can generate perfectly tailored, context-aware emails in any language, mimicking the tone and style of a specific executive with terrifying accuracy. ...