Anthropic has rewritten the biology classifier inside Claude Fable 5, cutting biology-related fallbacks by 85 per cent. Everyday health, clinical, and educational queries now reach Fable 5 directly — while dual-use domains like virology and toxicology remain restricted.
Firebird established a major artificial intelligence computing hub in Armenia, leveraging NVIDIA accelerated computing and Dell Technologies infrastructure to create the Commonwealth of Independent States' largest AI factory. The facility signals accelerating regional consolidation of AI infrastructure and represents a strategic expansion of cloud compute capacity in Eastern Europe.
ByteDance is reportedly training an AI model with as many as 10 trillion parameters — more than three times China's current largest open model and potentially larger than Anthropic's Mythos 5 — as Chinese tech giants escalate their push to reach the frontier of global AI capability.
Alibaba has released Qwen3.8-Max — 2.4 trillion parameters, 95B active — and will open-source its weights for the first time. Benchmarked across three real autonomous tasks: 16 days of self-evolving code generation (265 commits, zero human help), autonomous reproduction and improvement of an AI research paper (+2.71pts on AIME24), and beating 87% of 526 human teams in a 24-hour competition.
AWS, Cursor, GitHub, Microsoft, OpenAI, and Vercel have jointly backed "Agent Plugins" — an open specification that collapses the fragmented world of agent packaging into a single plugin.json manifest, ending the era of platform-specific glue and marking what may be the TCP/IP moment for autonomous AI agents.
AMD reported record Q2 2026 revenue of $11.5 billion — up 50% year-over-year — with Data Center revenue doubling to $6.7 billion on the back of Instinct GPU deployments at Anthropic, Meta, Microsoft, and OpenAI, and the launch of the Helios rackscale AI server, MI400 Series GPUs, 6th Gen EPYC, and the ROCm.ai developer platform.
Nvidia Omniverse and over 200 organizations are rallying around open-weights AI architectures to democratize physical AI capabilities across enterprise sectors. The shift signals a fundamental realignment in how AI infrastructure vendors and adopters approach model transparency, ecosystem development, and competitive positioning in robotics and autonomous systems.
Meta has launched browser-based calling on WhatsApp Web alongside call transfer between devices, a waiting room for group calls, instant HD video, and noise suppression — five features that close the gap with Teams and Zoom while positioning WhatsApp as a serious enterprise communication layer in markets where app downloads are restricted.
OpenAI has introduced dedicated ChatGPT plugins for K–12 teachers, college faculty, and university students, bundling role-specific workflows, curriculum-aware context, and institution-managed privacy controls into a single starting point — and formalising a five-year partnership with the American Federation of Teachers to train 400,000 US educators.
DataRobot and Dell Technologies executives outline how existing predictive AI infrastructure can serve as a foundation for autonomous agent systems, enabling enterprises to extract agentic AI value from established model portfolios without complete technology rebuilds.
Anthropic has appointed Mariano-Florentino (Tino) Cuéllar — former President of the Carnegie Endowment for International Peace, California Supreme Court Justice, and Stanford Law professor — as its first Chief Global Affairs Officer, formalising the company's approach to AI governance and international policy engagement.
NVIDIA's Alpamayo 2 Super is now available commercially under an open licence — and it ranks first on every autonomous driving benchmark NVIDIA evaluated, outscoring GPT-4o by 23.2 points and Gemini 2.5 Pro by 15.1 points on LingoQA. It is the clearest signal yet that NVIDIA intends to own the AV software layer as completely as it owns the GPU compute underneath it.
The Open Secure AI Alliance, comprising over 120 organizations, is developing new cybersecurity guidelines for agentic AI systems. The Linux Foundation has released a Request for Comments on Shared AI Findings Exchange, establishing frameworks for security transparency and threat intelligence sharing across the enterprise AI ecosystem.
As artificial intelligence workloads expand context windows and dataset requirements beyond traditional system memory limits, enterprise storage infrastructure is undergoing fundamental redesign. According to NVIDIA's analysis, the convergence of massive AI datasets and memory constraints is shifting how organizations architect data pipelines, requiring integrated solutions that combine capacity, security, and computational efficiency.
Unity has made its AI Gateway generally available, enabling enterprises to integrate multiple AI models through a unified API layer. The platform addresses fragmentation in ML model deployment and governance across distributed development environments.
Runware has built a vertically integrated AI inference platform — custom hardware, the Sonic Inference Engine®, and a single unified API — claiming up to 90% cost savings on serverless inference. With $63M raised, 300M end users, and 10B requests processed, it is quietly reshaping how the development layer accesses AI compute.
Etched has raised $300M at a $10.3B valuation, shipped A0 silicon from TSMC's N4P node, and landed over $1B in customer contracts — all before most of the industry noticed. Its bet: purpose-built frontier inference clusters will outperform GPUs on every metric that matters for production AI.
Meta and BlackRock have struck a joint venture to develop a one-gigawatt AI data center campus in El Paso, Texas — a $14 billion project in which BlackRock takes an 80% ownership stake, marking a structural shift in how Big Tech finances the AI infrastructure arms race.
OpenAI has published a detailed public rebuttal to Apple's trade-secret lawsuit — releasing private iMessages and emails that it says contradict Apple's own court filings and reveal that Apple employees asked the accused engineer to transfer files after he had already left the company.
Smarter, operating within the Salesforce ecosystem, has introduced advanced merchandising capabilities and platform extensibility features designed to address buyer search friction in B2B digital storefronts. The updates target the core challenge of product discoverability that drives cart abandonment across enterprise commerce platforms.
Billions are being poured into AI-powered drug discovery, yet the pipeline of AI-attributed medicines remains thin. The bottleneck isn't model capability — it's the quality of biological training data. A reproducibility crisis decades in the making is now directly limiting what AI can deliver.
DeepSeek's V4-Flash-0731 undercuts GPT-4o on inference cost by roughly 18x, ships with a 1M-token context window, and arrives with significantly enhanced agentic capabilities — making it the most cost-efficient option for organisations scaling autonomous AI workflows at volume.
Oracle's new AI-Native Builder Experience unifies no-code, low-code, and pro-code development inside Oracle Fusion Cloud Applications, enabling teams to build outcome-driven Fusion Agentic Applications — AI agent teams that reason, coordinate, and execute work directly against live enterprise data, workflows, and approvals.
In a strategic post authored by CFO Sarah Friar, OpenAI reveals that Codex now drives 99.8% of its weekly output tokens, ChatGPT has crossed one billion active users, and GPT‑5.6 Sol's ARC-AGI-3 score tripled through system optimisation alone — all pointing toward a model of intelligence as infrastructure, not product.