AI news digest — April 13, 2026
10 items, each with its source.
OpenAI Unveils GPT-5.5 With Continuous Real-Time Learning
OpenAI has officially announced GPT-5.5, introducing continuous real-time learning capabilities without requiring full model retraining. This breakthrough allows the model to update its knowledge base dynamically, significantly reducing hallucination rates in fast-changing domains.
openai.comEU AI Act Enforcement Begins For High-Risk Systems
The European Union has officially begun enforcing the strictest provisions of the AI Act for high-risk foundation models. Companies failing to comply with the new transparency and risk-management mandates now face fines of up to 7% of their global revenue.
reuters.comDeepMind Achieves Breakthrough in Infinite Context Window Scaling
Google DeepMind researchers published a new paper detailing a novel attention mechanism that effectively allows for infinite context windows. The technique reduces memory complexity from quadratic to linear, enabling LLMs to process entire libraries of text in seconds.
arxiv.orgAnthropic Releases Claude 4 Opus With Autonomous Agentic Workflows
Anthropic has launched Claude 4 Opus, featuring native support for autonomous agentic workflows. The model can now independently plan, execute, and verify complex multi-step tasks across various software applications without human intervention.
anthropic.comFigure 03 Humanoid Deployed in BMW Manufacturing Lines
Robotics startup Figure has officially deployed its third-generation humanoid robot, Figure 03, into active production lines at BMW's Spartanburg plant. The robots are performing complex, multi-step assembly tasks entirely autonomously using end-to-end neural networks.
techcrunch.comNVIDIA Unveils Rubin Architecture GPUs For Edge Inference
NVIDIA has announced its highly anticipated Rubin architecture GPUs, specifically designed to optimize edge AI inference. The new chips promise a 10x improvement in energy efficiency, making it possible to run massive parameter models on mobile devices.
nvidianews.nvidia.comMistral AI Secures $2 Billion Series D Funding Round
French AI startup Mistral AI has closed a $2 billion Series D funding round led by Andreessen Horowitz and the French government. The capital will be used to build a massive new supercomputer dedicated to training sovereign European open-source models.
bloomberg.comGlobal Watermarking Standard Adopted For AI-Generated Video
A coalition of major tech companies, including Meta, Google, and TikTok, has officially adopted a universal cryptographic watermarking standard for AI-generated video. The protocol embeds invisible, tamper-proof metadata to combat deepfakes ahead of upcoming global elections.
wired.comMicrosoft Azure Launches Quantum-Assisted AI Training Cluster
Microsoft has integrated its first commercial quantum computing nodes into Azure's AI training clusters. This hybrid classical-quantum infrastructure is designed to accelerate complex optimization problems in machine learning, drastically cutting down training times.
blogs.microsoft.comMIT Researchers Achieve Lossless 1-Bit LLM Quantization
A team at MIT has successfully demonstrated a lossless 1-bit quantization technique for large language models. This research milestone allows a 70-billion parameter model to run locally on standard consumer hardware without any degradation in reasoning capabilities.
news.mit.eduFeed, daily deep-dive and bytes — readable offline, with push alerts for the topics you follow.