Rushi Balapure

PICT Cyber Cell · PICT Pune, India · May 2026

Hands-on LLM Security Workshop

A hands-on PICT Cyber Cell workshop covering prompt injection, RAG attacks, excessive agency, Garak, Promptfoo, and defensive LLM system design, with public exercises and source code.

  • LLM Security
  • Gen-AI Red Teaming
  • Nvidia-Garak
LinkedIn workshop recap Open-source workshop materials

Returned to Pune Institute of Computer Technology as a guest speaker for a hands-on workshop hosted by the PICT Cyber Cell.

What the workshop covered

  • Threat foundations: OWASP Top 10 for LLM Applications, MITRE ATLAS, and how language-model threat boundaries differ from traditional software.
  • Automated testing: Vulnerability scanning and red teaming with tools including Garak and Promptfoo.
  • Prompt injection: Direct attacks and indirect attacks delivered through data retrieved by RAG systems.
  • Excessive agency: Exploiting autonomous agents and unsafe tool access.
  • Defensive design: Context minimization, action-selector patterns, and why prompt-only defenses are insufficient.

Participants ran the exercises locally with Python, Ollama, and an intentionally vulnerable model. The complete guide, lesson material, code samples, offensive exercises, and secure-agent examples are available in the public workshop repository.