Mitigating Indirect Prompt Injection Attacks in Autonomous AI Agents Using Context-Aware Token Sanitization

A defence against indirect prompt injection attacks in autonomous AI agents using context-aware token sanitisation.

A defence against indirect prompt injection attacks in autonomous AI agents using context-aware token sanitisation. This reference project walks through the whole job, from the problem and the design to the build and the evaluation, so that a developer can study how it works, run it and extend it.

It suits learners who want practical security experience: understanding the threat, building the defence or detector, testing it against realistic attacks and reporting the results.

The main parts are cooperating agents with defined roles and tools, a realistic threat model, the detection or protection mechanism itself, testing against sample attacks, and reporting and alerting of findings. The reference build uses LLM APIs, NLTK, Python, scikit-learn, and Wireshark, with a clear project structure, setup instructions and sample data so that it can be run and extended.