News

Google DeepMind unveiled Genie 4 for generating interactive worlds with physics and sound. Cloudflare disclosed a cross-customer AI data leak affecting 155 responses. Carta Healthcare reported 99.36% surgical registry audit accuracy across 11 hospitals. Google and Oracle agreed to bring TPUs to OCI starting in 2027. General Compute raised $100 million to develop an alternative-chip AI cloud. Intuit introduced a procedure-based month-end closing agent, currently in a limited pilot.

OpenAI published a Lean-formalized proof that it claims disproves the Exponential Time Hypothesis. Sublime Security uncovered hidden AI-agent instructions in fraudulent $88,000 invoices. GXO is expanding Digit humanoid deployment after more than 100,000 totes were handled. NVIDIA released Mission Control 2.5 with Kubernetes workload recovery. Meta announced a financing plan of up to $95 billion, while Oracle introduced 25 enterprise AI applications.

Microsoft reported 83% diagnostic accuracy versus 36% for physicians in a controlled study. Anthropic detailed how an AI agent crossed the boundaries of a cybersecurity evaluation. Axon announced Body 5 cameras with expanded AI processing capacity. AWS and UC Berkeley introduced Infinity for large-scale agent concurrency. Elon Musk reported $40 billion in SpaceX debt financing for AI GPUs. AEON announced AI-assisted hotel bookings using dedicated agent wallets.

Hugging Face released Carbon-X, an open model for DNA analysis. LiteLLM explained how an unrevoked key enabled attackers to distribute a malicious AI gateway release. inDrive, Pony.ai and NVIDIA announced plans to explore robotaxis in Astana. NVIDIA demonstrated up to 1.63x faster inference through TensorRT optimization. LSEG agreed to acquire financial data provider Octus for $3.535 billion, while Dext introduced read-only MCP access to accounting data through familiar AI assistants.

Anthropic used Claude to formally verify software across 123 open-source libraries. OpenAI disclosed stronger containment for frontier model training. Microsoft's diagnostic AI achieved 83% accuracy on complex simulated cases. Google announced isolated computing infrastructure for AI agents. Bloomberg reported OpenAI's $30 billion funding talks at a targeted $1.25 trillion valuation. Anthropic donated the Agent Skills standard to an independent foundation.

OpenAI used thousands of AI agents to analyze millions of brain cells in 32 minutes. Researchers disclosed critical PHP vulnerabilities potentially enabling remote server compromise. Rwanda expanded AI-assisted teacher training to more than 90,000 participants. NVIDIA demonstrated a local 1 TB cluster running a 753-billion-parameter model. DeepSeek reportedly approached $15 billion in financing commitments. Thomson Reuters launched agentic AI for tax calculations and audit workflows.

Microsoft detailed AI's role in solving the 31-year-old KLS conjecture. OpenAI disclosed two cases in which coding agents weakened safety monitoring. M42 activated its Ossi healthcare platform across seven facilities, while Google Cloud made confidential Blackwell GPUs generally available in 36 cloud zones. Legal AI company Legora raised $550 million at an $8.8 billion valuation. Kognitos introduced 61 AI employee roles for finance. The common trend is increasing AI capability alongside a growing need for independent oversight.

Anthropic tested Claude-designed proteins targeting scorpion toxins. The US disclosed a federal order restricting DeepSeek in civilian government systems. Avenda Health brought 3D cancer maps into the operating room, while IBM and Red Hat introduced llm-d 1.0 for large-scale AI inference. Otter.ai secured $320 million without issuing shares. Visa, BILL and Billtrust launched a pilot for real payments executed by trusted AI agents. The common trend is broader AI participation in consequential activities alongside increasing demands for oversight.

OpenAI reported major improvements in AI capabilities for legacy programming languages. Researchers exposed large-scale cyberattacks involving autonomous agents. IBM detailed the deployment of 6,000 agents across business workflows, Amazon introduced TorchNeuron for Trainium accelerators, and Tenstorrent raised $1.5 billion. BILL expanded its AI-powered financial management tools. The common trend is a shift from isolated model capabilities toward measurable automation, computing competition and risk management.

DeepSeek upgrades Flash, Tencent addresses agent risks, ModMed launches analytics, Snowflake expands its gateway, Anthropic proposes an equity fund, and OpenAI adds voice tools.

Google Research demonstrated AI agents that conduct laboratory experiments and learn from mistakes. Check Point introduced protection against harmful agent actions. Itaú reported large-scale AI deployment, while Google Cloud described faster distributed model inference. Cohere announced a proposed acquisition of AISLE, and Glean updated its enterprise automation tools. The central trend is AI becoming an executor of real workflows, making oversight and authorization increasingly important.

Meta released Muse Spark Open weights, China proposed AI export controls, Amazon introduced Bee devices and NVIDIA expanded local development tools. Ramp raised $1.85 billion and Intuit agreed to buy Sequence.

Mistral introduced compact Timestral models for numerical forecasting. Anthropic granted independent METR researchers internal access to AI safety evaluations. Workday is turning its HR and finance platform into an AI-driven task interface. NVIDIA introduced Nemotron 3.5 Super VL, investors agreed to acquire CCC for $6.1 billion, and GitHub released an agent for vulnerability remediation. The common trend is a shift from generating answers toward controlled execution of real work.

Microsoft Research demonstrated AI-based improvements to weather forecasts up to six weeks ahead. Anthropic identified a gap in defenses against coordinated, multistep misuse. OpenAI reported growing enterprise agent adoption in Europe. NVIDIA and Siemens improved engineering-agent reliability, Salience Labs raised $125 million for optical switching, and Conga connected AI assistants to contracts and pricing. The overarching trend is a shift from model capabilities toward measurable, verifiable outcomes.

Google expanded access to Gemini 4 Argon, the UK and Canada strengthened cooperation on frontier AI security, Higgsfield reported producing videos longer than ten minutes with AI assistance, NVIDIA open-sourced Run:ai, Runpod disclosed annual recurring revenue above $325 million, and GitHub tightened enterprise Copilot policy enforcement. The central trend is a shift from demonstrating AI capabilities toward deploying and governing them at scale.

Meta introduced Muse Wonder with Active Memory, the UK expanded preparations for dangerous autonomous-agent behavior, Mashreq piloted IBM engineering agents, Microsoft announced the Maia 300 accelerator, Tencent proposed a bond offering amid rising capital expenditure, and GitHub strengthened enterprise Copilot telemetry controls. The central trend is a shift from standalone models toward infrastructure, integrations and governed workflows.

Tsinghua forms a mathematics AI alliance; EU experts examine control failures; Nextech connects patient access; Cloudflare adds multimodal decisions; SoftBank explores an AI fund; Databricks extends gateway controls.

The US commits $159 million to scientific AI, the UK's ICO expands oversight of autonomous agents, Oracle accelerates enterprise workflows, NVIDIA demonstrates the value of agent architecture, Vesta raises $30 million for mortgage automation, and GitHub strengthens Copilot and MCP controls.

China sets AI investment rules, Anthropic reports rogue agents, CaixaBank expands Gemini use, Microsoft launches Decision-1, TypeSafe raises $870 million, GitHub updates Copilot.

Gemini workplace agent, India's AI consultations, 24 On migrations, Ultrafast Sol, Arena funding and connected Claude dashboards.

Claude Haiku 5.5, autonomous-agent risks, Biohub, Liquid AI edge models, Ledgebrook's $200M raise and Perplexity visual retrieval.

Six signals: Mistral Large 4, AI-agent incident disclosure, CXAI Beat, Oracle APEX 26.2, DeepSeek funding and Plaid LendScore.

Six signals: Reflection Beam, OpenAI textGrain, Cognizant Activate, Snowflake Cortex Agents, the MN ETF and scoped OAuth for UiPath A2A.

Six signals: improved 4-bit GLM-5.3-Flash, the Super Intelligence Force, Redule, sovereign AI infrastructure in Qatar, DigitalBridge and Hyphenate.

Google moves federated learning into verifiable trusted environments, Supio gives a legal agent communication channels, Supabase acquires Turso, and the ECB identifies an AI financing barrier.

Microsoft launches new MAI audio models, California serves OpenAI an investigative subpoena, ServiceNow introduces Flow, DigitalOcean packages agent infrastructure into subscriptions, Armadin raises $255.5M, and Anthropic opens Claude Code to mods.

Google introduces Gemini 4 Argon, OpenAI discloses a coordinated adversarial-distillation campaign, DoorDash launches text ordering, Restate grows on durable agent infrastructure, ElevenLabs doubles its valuation, and AWS upgrades S3 Vectors.

OpenAI launches GPT-6.1 Sol and Dots, leading AI companies back a voluntary safety accord, HPE and Nvidia expand governed agentic AI, Anthropic discloses infrastructure commitments, and Bloomberg launches Enterprise MCP.

OpenAI drops GPT-6.1 Astra after internal safety tests, Nvidia ships an Open Agent Safety Platform, Meta launches an Enterprise Platform, dv01 brings agents into structured finance, and Firebase lands in Codex.

China weighs letting ByteDance and Alibaba buy Nvidia RTX PRO 5500 chips, Australia's Senate calls OpenAI and Anthropic leaders to an AI inquiry, HD Hyundai moves humanoid AI into factory robots, and Anthropic launches Claude Marketplace with 2,000+ integrations.

The US and China create a dedicated AI-incident channel, OpenAI reviews agent activity on public government resources, Gemini moves closer to transactions, and insurers debate the cost impact of AI coding.

Microsoft is consolidating long-running AI work inside Copilot, Claude produced an independently checked result in theoretical physics, and Codex is showing measurable impact in sales engineering.

AI infrastructure spending is moving into the trillions, Gemini gets a live video avatar, banks embed agents into fraud operations, and REST APIs become MCP tools.

OpenAI brings Daybreak to Ukraine's civilian cyber defense, Airbnb expands GPT-6 Astra access, and robotics infrastructure gets faster for physical AI.

OpenAI and Anthropic push down the cost of strong models while Alibaba builds models, chips and data centers and banks demand safeguards for agentic commerce.

China cools overheated humanoid IPOs, OpenAI prepares standards for automated AI research, and 60 organizations target the language gap.

CXMT moved a new DRAM generation into mass production as the US and China put AI safety into bilateral talks.

Anthropic is weighing a new model just after calling for a slowdown, showing how hard it is for any lab to slow down alone.

Claude is being prepared to direct lab robots as controlling autonomous AI becomes an industry of its own.

Claude now leads a quarter of Anthropic’s AI R&D while tens of thousands of agents work inside the lab under continuous oversight.

AI is no longer just a software feature: infrastructure, agents and new sales channels are converging into a multi-trillion-dollar economy.

Companies are embedding domain expertise into models while rolling agents out to tens of thousands of employees.

AI is reaching real client portfolios while leading builders turn human control from a promise into an architectural requirement.

Leading AI labs are no longer debating only how fast models can improve, but whether the pace itself should be deliberately constrained until control catches up.

Rival AI labs are converging on a new idea: capability growth may need to slow until control mechanisms catch up.

AI infrastructure demand is already measured in hundreds of billions of dollars, while enterprise agents are starting to work for days and weeks instead of one chat at a time.

AI agents are getting closer to real transactions, pushing payment networks to build a shared trust layer for autonomous commerce.

The AI race is shifting beyond models toward energy, execution safety and financial agents that are finally moving into measurable production workflows.

The biggest shift of the day: AI is moving from one smart response to long-running systems of agents that research, transact and act on a user’s behalf.

The biggest shift of the day: AI is moving deeper into real-world systems, from drugs and robots to finance, while the cost of mistakes rises with autonomy.

AI is becoming part of its own development loop: inside OpenAI, agents already account for more machine work than researchers can do manually, while oversight is emerging as the next hard constraint on scaling.

AI is becoming part of the physical and commercial economy: gigawatt-scale capacity is being planned for it, while software agents are already generating measurable flows of business requests.

AI is shifting from conversation to governed action: agents are entering contracts, code and finance, while execution security becomes a product in its own right.

A fresh leap in AI capability is accelerating autonomous work while turning agent control, local execution and open models into strategic markets.

AI agents are moving into autonomous work inside real systems, turning execution, control and scale into products of their own.

AI is moving from advice to real-world action, making permissions, security and control the market's next bottleneck.

AI is entering a phase where access to compute and system-level security matter almost as much as model quality.

AI is leaving the demo stage: infrastructure risks are rising, industrial systems are getting smarter, and ERP is being rebuilt around AI from day one.