Breaking Gemini 2.5 Pro Multimodal in audio and vision mode

We just published our red team report on Google Gemini 2.5. It’s the first in-depth security review of Gemini’s multimodal models at enterprise scale.

Found some serious issues with image, audio, and text-based attacks, especially important as more teams plug these models into agents and workflows

Link here if you want to check it out or share:

Happy to do a quick recap if helpful!