← Back to projects & research
Prompt injection research2026
Prompt injection research
Authored a published survey of prompt injection attacks and built a Python/Flask detection classifier.
Outcome
Published survey plus a working classifier validated by 83 passing pytest tests.
The problem
Large language models remain vulnerable to prompt injection, and the attack surface had no consolidated taxonomy to defend against.
What I built and owned
A published foundation
“Hijacking the Prompt: A Survey of Prompt Injection Attacks, Detection, and Defense in Large Language Models” was published via ODU Digital Commons under the Commonwealth Cyber Initiative (COVA CCI).
From taxonomy to working software
A Python and Flask classifier matches 13 prompt injection signatures against a six-category taxonomy. The project includes 83 passing pytest tests.