tuandunghcmut/trustsight-synthetic-suite

TrustSight Synthetic Visual Suite: In-Contract Misuse & Typographic Injection Benchmark 1. Overview & Research Motivation TrustSight Synthetic Visual Suite is a controlled, mathematically balanced benchmark designed to evaluate multimodal agents (Vision-Language Models with tool-calling capabilities) against Typographic Visual Prompt Injections (VPI), subtle In-Contract Steering, and Out-of-Contract Exfiltration. Unlike unconstrained web scraping datasets where… See the full description on the dataset page: https://huggingface.co/datasets/tuandunghcmut/trustsight-synthetic-suite.

<iframe src="https://silytics.si/embed/item/b6b8d88b-56c2-4ffd-9fdc-f4c56aa5a3e0" width="420" height="190" style="border:0" loading="lazy" title="Silytics entry"></iframe>

Silytics shows public metadata from Hugging Face and links to the original. Information may be incomplete or out of date. Last refreshed 17 October 2026. Licenses and terms of the original apply to the work itself. Not financial, legal or professional advice.