AI Digest

Google Unveils Gemini 3 with Multimodal Reasoning Breakthrough
Google DeepMind released Gemini 3, its latest flagship model, featuring native video understanding and real-time multimodal reasoning. The model can watch a live video feed, answer questions about physical objects, and suggest actions in real time, marking a significant step beyond text-and-image reasoning. Early benchmarks show a 40% improvement on complex spatial reasoning tasks over Gemini 2.

EU Parliament Passes AI Liability Directive
The European Parliament approved the long-debated AI Liability Directive, imposing a strict liability framework for high-risk AI systems. Under the new rules, companies deploying AI in critical sectors like healthcare and transportation must prove their systems were not defective in the event of harm. The directive is expected to take effect in 2027, with fines reaching up to 6% of global annual turnover.

OpenAI Launches GPT-5 Turbo for Enterprise Customers
OpenAI released GPT-5 Turbo, a faster and cheaper version of its flagship model designed for high-volume business use. The model achieves a 3x reduction in inference cost while maintaining near-identical performance on standard benchmarks. OpenAI also introduced a new "context caching" feature that cuts API costs for long-running conversations by up to 80%.

Apple Acquires AI Chip Startup NeuReality for $1.2 Billion
Apple confirmed its acquisition of Israeli AI chip startup NeuReality, a move aimed at reducing reliance on Nvidia for on-device AI processing. NeuReality's specialized inference chips are designed for low-power, high-throughput edge computing, which Apple plans to integrate into future iPhones and Vision Pro headsets. The deal is expected to close by Q3 2026.

Meta Shows Off AI-Powered Smart Glasses with Live Translation
Meta demonstrated a new version of its Ray-Ban smart glasses featuring real-time, hands-free language translation for 25 languages. The glasses use a combination of on-device processing and cloud assistance to display translated subtitles directly in the user's field of view. Meta says the feature will launch as a beta update next month.

Chinese Researchers Claim Quantum Advantage in Drug Discovery
A team from the University of Science and Technology of China published a paper claiming a quantum computer simulated a complex protein folding problem 1,000 times faster than the best classical supercomputer. While the problem is highly specific, researchers say it is the first practical demonstration of quantum advantage in a real-world pharmaceutical application. The work has not yet been peer-reviewed.

Microsoft Shuts Down AI-Powered Windows 12 "Copilot+" Feature After Privacy Backlash
Microsoft temporarily disabled "Recall," a flagship AI feature in Windows 12 that continuously captured screenshots of user activity to enable search and task automation. The decision came after security researchers demonstrated that the locally stored screenshot database was accessible to malware and could be exfiltrated without user permission. Microsoft says it will redesign the feature with hardware-level encryption before re-enabling it.

Nvidia Unveils Blackwell Ultra GPU for AI Training
Nvidia announced the Blackwell Ultra GPU, its next-generation AI training chip, claiming a 4x performance improvement over the Hopper architecture. The chip features 288 GB of HBM4 memory and a new interconnect that allows seamless scaling to clusters of over 100,000 GPUs. Major cloud providers have already placed pre-orders, with shipments expected in early 2027.

Editor's take: Today's news reveals a clear tension between rapid AI deployment and the growing demand for accountability. From Microsoft's privacy misstep to the EU's strict new liability rules, the industry is being forced to confront the real-world consequences of its products. Meanwhile, hardware and model releases from Google, OpenAI, and Nvidia show that the pace of capability improvement shows no sign of slowing. The divergence between what is technically possible and what is socially acceptable is the defining story of this moment.