NeuralStackApple Integrates On-Device AI Features into iOS 20 Beta
Apple seeded iOS 20 beta on July 14 with on-device large language model capabilities. The update runs a 3 billion parameter model locally for summarization and keyboard prediction. It reduces cloud API calls by 70 percent for supported tasks on A19 chips.
Google Open Sources JAX 1.0 with Distributed Training Primitives
Google released JAX 1.0 on July 15 with built-in distributed training primitives and automatic sharding. The update supports training runs across 16,000 TPU v6 chips without custom code. Early benchmarks show 2.4 times faster convergence than JAX 0.4 on large language models.
EU Commission Proposes AI Liability Directive Update on July 21
The European Commission proposed amendments to the AI Liability Directive on July 21. The draft introduces strict liability for high-risk AI systems and mandatory insurance requirements. Member states must transpose the rules by December 2027 if approved.
Stanford Researchers Achieve 98 Percent Accuracy on ARC-AGI Benchmark
Stanford University researchers published a new architecture on July 16 that scores 98 percent on the ARC-AGI benchmark. The system uses dynamic program synthesis and requires only 400 training examples. The result surpasses previous state-of-the-art by 19 percentage points.
Amazon Web Services Launches Bedrock 2 with Agent Builder
AWS released Bedrock 2 on July 20 adding a visual agent builder and 12 new foundation models. The update includes native support for multi-agent orchestration and real-time tool calling. Early customers report 60 percent faster development cycles for internal automation projects.
LockBit Claims Breach of 3.1 Million Bank of America Records
LockBit published 3.1 million Bank of America customer records on July 17 claiming a May 2026 intrusion. The data includes account numbers and partial Social Security numbers. Bank of America stated it detected and contained the intrusion on June 4.
Cohere Raises 900 Million Series D at 9.2 Billion Valuation
Cohere closed a 900 million dollar Series D round on July 19 at a 9.2 billion valuation. Inovia Capital and Tiger Global led the round with participation from existing investors. The capital will fund expansion of its enterprise RAG platform across Europe and Asia.
Meta Unveils Llama 4 with 10M Token Context on July 18
Meta released Llama 4 on July 18 featuring a 10 million token context window and native video processing. The model scores 94.2 on MMLU and runs inference 3.8 times faster than Llama 3.1 405B. It positions Meta to challenge closed models in enterprise deployments.
Anthropic Publishes Constitutional AI 2.0 Research Paper
Anthropic released Constitutional AI 2.0 on July 14 2026 improving harmlessness scores by 34 percent over prior methods. The technique uses 180 principle-based rules evaluated by a separate critique model. The paper reports results across 12 model sizes up to 200 billion parameters.
Microsoft Adds 45 Billion Dollars to AI Infrastructure Budget for 2027
Microsoft increased its 2027 AI infrastructure budget to 95 billion dollars on July 21 2026. The additional 45 billion dollars will fund new data centers in Texas and Arizona. Azure AI revenue grew 78 percent year-over-year in Q2.
Hugging Face Launches Transformers 5.0 with Native Multimodal Support
Hugging Face shipped Transformers 5.0 on July 15 2026 adding first-class multimodal model support and automatic pipeline optimization. The update includes 340 new model architectures and 2x faster inference on consumer GPUs. Over 1.2 million developers have already upgraded.
US Senate Passes Bipartisan AI Safety Certification Act
The US Senate passed the AI Safety Certification Act on July 16 2026 requiring third-party audits for models above 10^26 FLOPs. The bill mandates annual safety reports and incident disclosure within 72 hours. It now moves to the House for reconciliation.