AI News Briefing — June 11, 2026

A senior Ukrainian defense industry figure told New Scientist that fully autonomous drones programmed to destroy anything in a designated area produced confirmed human casualties in a test about two years ago. It is described as the first known case of fully autonomous AI weapons killing human soldiers.

Google DeepMind released DiffusionGemma, a new Gemma 4 open model that generates text via parallel diffusion-style denoising rather than autoregressive token-by-token decoding, claiming up to 4x faster local inference. It is positioned as an experimental tool for developers running on consumer or workstation GPUs.

Continuing our coverage of the German AI Overviews ruling, A German court issued a preliminary ruling that Google is liable for false statements generated in its AI Overviews after the feature labeled publishers as scams and failed to correct the output following a cease-and-desist. The decision could set precedent affecting all AI search engines and chatbots that misattribute or fabricate claims about sources.

The Decoder reports on Google's DiffusionGemma, a 26-billion-parameter open model that generates text through diffusion rather than autoregression, hitting roughly 1,000 tokens per second on an H100—about 4x faster than comparable models—at the cost of lower output quality. Google positions it as an experimental developer tool.

OpenAI is reportedly negotiating to lease a planned 10-gigawatt data center in Ohio, with potential financial backing from Nvidia. The deal would be OpenAI's largest compute facility yet.

Amazon borrowed $17.5 billion from banks shortly after a bond sale to fund continued AI spending, underscoring the heavy debt loads tech firms are taking on for the AI arms race. The move highlights mounting financial pressure across hyperscalers.