Microsoft

A one-prompt attack that breaks LLM safety alignment

As LLMs and diffusion models power more applications, their safety alignment becomes critical. Our research shows that even minimal downstream fine‑tuning can weaken safeguards, raising a key question ...
GitHub

Dual-Flow

[2025-10-18]: The code and models have been released! We have improved the training resolution of the models and achieved better results compared to the original paper. Both old weights (for ...
IEEE

Adversarial Diffusion Model: Generating High Quality and Undetectable Images from Scratch

Abstract: Diffusion models have made tremendous progress in generating visually realistic images. However, these images are statistically different from the real images, which could be accurately ...
The Verge

X claims it has stopped Grok from undressing people, but of course it hasn’t

It’s still easy to get Grok to edit photographs of real people into sexualized poses, despite X’s updated restrictions. It’s still easy to get Grok to edit photographs of real people into sexualized ...
SlashGear

China's Wind Farms Are Doing A Lot More Than Generating Electricity

Nations around the world are dedicating a tremendous amount of resources to projects that can help reduce or restore some of the damage caused by climate change. One significant change that's taking ...
Wired

Grok Is Generating Sexual Content Far More Graphic Than What's on X

This story contains descriptions of explicit sexual content and sexual violence. Elon Musk’s Grok chatbot has drawn outrage and calls for investigation after being used to flood X with “undressed” ...
Wall Street Journal

Mexico Is Acting Like an Adversary to the U.S.

Free Expression: In reality the Maduros' removal from Venezuela was a superbly executed act of strategic opportunism that removed a vexing enemy, and it's a myth to suggest 'Operation Absolute Resolve ...
IEEE

IlluAttack: Generating Adversarial Examples via Illumination Transformation for Face Anti-Spoofing

Abstract: Existing adversarial attacks for face anti-spoofing predominantly assume that the model parameters are fully known, and often overlook the transferability of adversarial examples across ...
POWER Magazine

Framatome Selected to Deliver Critical Digital Control Upgrades at Columbia Generating Station

Framatome said it has been awarded a contract to upgrade several digital control systems at Energy Northwest’s Columbia Generating Station. The scope of the project includes system design, engineering ...
GitHub

Failed to generate adversarial image: Script execution failed:

127.0.0.1 - - [12/Nov/2025 16:07:06] "GET /api/decoy-images HTTP/1.1" 200 - 127.0.0.1 - - [12/Nov/2025 16:07:06] "GET /api/demo-images HTTP/1.1" 200 - 127.0.0.1 ...
Scientific Research Publishing

Zhang, Y., Jin, R. and Zhou, Z.-H. (2018) Understanding Adversarial Examples Systematically: Exploring Data Perturbations in Cybersecurity.

ABSTRACT: In recent times, the World has experienced the impact of artificial intelligence (AI) and machine learning (ML) on the digital ecosystem, organizations now face both unprecedented ...