Live intelligence

Latest AI updates

2 updates
Research

What Survives the Shift from Text to Pixels? Conditional Protection against Prompt Injection in Image-Rendered Documents

This arXiv paper studies indirect prompt injection when multimodal language models answer questions over external documents, comparing text and page-image versions of 191 English QA sources across six models. Rendering untrusted content as images reduces marker-appending and answer-formatting execution for every model, but Qwen3-VL-8B still executes image-borne answer replacement at 77.96%. Protection varies by target type, rendering settings, wording, and repetition, and an eight-candidate attack search largely closes the image deficit for Qwen and Gemini but not GPT-5-nano. Clean-question utility ranges from 93.31% to 101.61%, though Qwen drops to 61.99% when answers require combining two evidence spans.

Prompt injectionMultimodal language modelsAI security
Research Editor selected

Geometric Abstraction Reduces Multimodal Queries in Agentic Robot Policies

This paper proposes replacing step-by-step visual reasoning in multimodal robot policies with a 3D geometric abstraction that converts scene geometry into typed, evidence-aware symbolic references grounded from RGB-D data. The system refreshes metric values locally and returns compact textual checkpoints after each action, invoking direct visual reasoning only on demand. Experiments show improved multi-step manipulation completion while substantially reducing repeated multimodal queries.

Why it matters: Offers a concrete method to improve efficiency and reliability of agentic robot policies by using geometric abstraction, relevant to AI-driven robotics.

robot manipulationmultimodal language modelsgeometric reasoning