// Demos

Interactive demonstration of token counting and rate limiting strategies to defend against unbounded consumption attacks.

Explore benchmark results from TruthfulQA and HELM, comparing how different models perform on factual accuracy tasks.

Conceptual visualization of how malicious documents shift embedding space, causing RAG retrieval to surface attacker-controlled content.

Interactive demo with a hidden system prompt — try to extract it using prompt injection techniques.

Interactive demonstration of how an LLM agent's tool-calling behavior changes when given progressively more dangerous permissions.

Side-by-side comparison of raw LLM output rendered as HTML versus properly sanitized output, demonstrating XSS prevention.

Conceptual demonstration of how activation patterns change when trigger tokens are present in a backdoored model.

Visual exploration of the dependency chain in a typical LLM deployment — from base model to adapters, tokenizers, and plugins.

Side-by-side comparison showing how subtle prompt modifications can turn benign queries into data extraction attempts.

Experiment with direct and indirect prompt injection payloads against a simulated LLM system prompt.