AI Digest

OpenAI Releases GPT-5 with Enhanced Reasoning and Multimodal Capabilities
OpenAI officially launched GPT-5 today, its latest flagship model, which demonstrates significantly improved reasoning in mathematics, coding, and complex planning. The model integrates native multimodal input, processing text, images, and audio simultaneously, and is available to ChatGPT Plus users and via API. Early benchmarks show GPT-5 outperforming GPT-4 by over 30% on several standardized reasoning tests.

Google Unveils Gemini 2.0 Pro with 2 Million Token Context Window
Google DeepMind released Gemini 2.0 Pro, featuring a 2 million-token context window—double the capacity of its predecessor. This allows the model to process entire books or extensive codebases in a single query. The model also introduces new agentic capabilities, including the ability to browse the web and execute code autonomously.

EU Passes Landmark AI Liability Directive
The European Parliament voted to approve the AI Liability Directive, creating a legal framework for holding AI developers and deployers accountable for harm caused by their systems. The directive introduces a "presumption of causality" in certain high-risk AI cases, shifting the burden of proof toward companies. It is expected to take effect in member states by early 2027.

Meta Releases Open-Source CodeGen Agent for Software Development
Meta open-sourced CodeGen Agent, a tool that autonomously writes, tests, and debugs code based on natural language descriptions. The agent is built on a fine-tuned Llama model and integrates directly with GitHub and VS Code. Early tests show it can resolve over 60% of standard software engineering tasks without human intervention.

Apple Announces On-Device AI Health Coach in iOS 20
Apple previewed a new AI-powered health coaching feature coming with iOS 20, which runs entirely on-device using the Neural Engine. The coach analyzes user activity, sleep, and biometric data from the Apple Watch to provide personalized wellness recommendations. Apple emphasized that all processing stays on the device, addressing privacy concerns.

Nvidia Reports Record Data Center Revenue Driven by Blackwell GPU Sales
Nvidia announced quarterly data center revenue of $45 billion, a 40% year-over-year increase, fueled by massive demand for its Blackwell architecture GPUs. CEO Jensen Huang attributed the growth to "the industrialization of AI" as hyperscalers and enterprises race to build out inference infrastructure. The company also unveiled a new networking chip designed to reduce latency in large-scale AI clusters.

China’s Ministry of Industry Bans Export of Advanced AI Training Algorithms
China’s Ministry of Industry and Information Technology imposed a ban on the export of proprietary AI training algorithms, citing national security concerns. The move specifically targets algorithms used for optimizing large language models and multimodal systems. The regulation will likely deepen the technological decoupling between China and Western AI firms.

Startup Anthropic Secures $10 Billion in New Funding for AI Safety Research
Anthropic announced a $10 billion funding round led by Spark Capital and existing backers, earmarked for expanding its safety research division and building a new supercomputer. The company plans to double its safety team to 1,000 researchers and focus on "alignment verification" techniques. The funding signals continued investor appetite for safety-focused AI development.

Editor's take: Today's news is dominated by a dual narrative of rapid model advancement and tightening regulatory frameworks. While GPT-5 and Gemini 2.0 Pro push the boundaries of what AI can do, the EU's liability directive and China's export ban mark a decisive turn toward governing these capabilities. The industry is entering a phase where technical progress and policy guardrails are accelerating in tandem, setting the stage for a more structured—but also more fragmented—global AI landscape.