AI news for Nov 16, 2026

24 stories

  1. PolicyOpenAI
    OpenAI partners with NATO on allied cyber defenses

    OpenAI says it is collaborating with NATO and cyber agencies from allied countries to support trusted defenders. The work focuses on helping those defenders locate vulnerabilities, test fixes, and protect essential systems, according to the company's announcement.

  2. BusinessOpenAI
    OpenAI and Chile's government partner on responsible AI adoption

    OpenAI announced a partnership with the Government of Chile aimed at supporting responsible AI adoption in the country. The agreement covers building public-sector skills and helping local startups develop AI applications. OpenAI did not say which tools or programs will be involved, or when the work begins.

  3. PolicyOpenAI
    OpenAI Adds Tools and Partnerships to Fight Online Child Abuse

    OpenAI announced new tools, practical guidance and partnerships intended to prevent, detect and respond to online child sexual exploitation and abuse. The company framed the effort as strengthening the fight against online child sexual abuse. The announcement did not name specific products, partners or availability details beyond describing the tools, guidance and partnerships.

  4. BusinessOpenAI
    LVMH's Media Division announces strategic partnership with OpenAI

    LVMH's Media Division and OpenAI have announced a strategic partnership. According to the announcement, the two organizations will work together to support journalism with AI and to make LVMH Media Division's publications more accessible through ChatGPT. The announcement does not specify a timeline, the titles involved or the terms of the arrangement.

  5. ProductsOpenAI
    OpenAI adds computer-use agents and healthcare plugins to ChatGPT for Healthcare

    OpenAI announced new capabilities for ChatGPT for Healthcare: computer-use agents, improved clinical answers, and healthcare plugins that work across healthcare software. The company says the additions are intended to help users get real work done.

  6. ProductsGoogle
    Google Maps adds Ask Maps itineraries and appointment booking

    Google announced new Ask Maps capabilities in Google Maps on November 16, 2026. Users can ask for a day-by-day trip itinerary with numbered map pins, share it, or export it to Google Docs on desktop, and adjust stops while traveling. A "Know before you go" panel on directions shows parking and charging spots, weather, potential delays and tolls. With new agentic capabilities, Ask Maps can also surface open appointment slots for beauty, pet care, auto and home repair, and call local businesses on the user's behalf. Users can search their saved lists for places with live availability. The updates roll out today on Android and iOS; itineraries and list search are available everywhere Ask Maps is offered, while route tips and appointment finding start in the U.S. Ask Maps also expands to 30 European countries, reaching more than 200 countries and territories.

  7. PolicyHacker News
    Researcher finds Windows 11 logs which coding agents you use

    A developer's reverse-engineering write-up reports that a Windows 11 component, MissionControlAggregator.dll, compares executable names such as codex.exe, claude.exe, copilot.exe and cursor.exe, along with developer tools like python.exe and gcc.exe, and records which ones were observed. The post says the component subscribes to existing Windows process and WSL events rather than scanning, keeps results locally, and emits a bitmap report through Event Tracing for Windows. On the author's ARM64 virtual machine, and on a clean US-region Windows 11 Home install, optional diagnostic data was enabled by default and the report was queued for upload to Microsoft. The author notes the feature was disabled by image default on the public ISO's baseline VM, and found no prompt text or source code in the report.

  8. Open sourceMicrosoft Research
    Microsoft Research and UC Berkeley release TokenEconomizer coding agent plugin

    Microsoft Research, with UC Berkeley and Microsoft Security, has released TokenEconomizer, an open-source plugin that uses logs from an AI coding agent’s earlier sessions in a repository to guide later ones. It adds short repository notes, prunes tools the agent is unlikely to need, and intervenes when an agent repeats a failure, without changing the agent or model. In tests on 277 SWE-Bench bug-fixing tasks across six repositories and five agent-model combinations, it cut average cost per task by 18.4–30.9% while success rates stayed broadly stable.

  9. Open sourcevLLM
    vLLM v0.34.0 adds 10 models and enables extensible KV cache by default

    It adds ten model architectures, among them Mistral Large 4, MiniMax-M3.1-Flash, NVIDIA Nemotron-3.5-Transcribe, Baidu Unlimited-OCR-2 and Liquid AI's LFM3 family. The extensible KV cache is now the default on CUDA and ROCm: cache size is derived from the memory left over after warmup and CUDA graph capture rather than an estimate, which the project says gives most models a larger cache. Model Runner V1 is removed, the default block size moves from 16 to 64, and ROCm 7.2 builds are no longer published, so existing deployments may need configuration changes.

  10. ProductsNVIDIA Newsroom
    NVIDIA Debuts Four-Node DGX Spark HPC Developer Kit for Researchers

    NVIDIA introduced the DGX Spark HPC Developer Kit at SC26, the supercomputing conference held this week in Chicago. The kit bundles four DGX Spark systems, each built on the GB10 Grace Blackwell Superchip, wired directly to its two neighbors in a ring over ConnectX-7 networking so no network switch is needed. At four nodes it offers 512GB of combined memory and 20TB of NVMe storage, and NVIDIA says models up to 700B parameters can run locally using NVFP4. A playbook for installing the NVIDIA HPC SDK is coming soon.

  11. ProductsNVIDIA Newsroom
    NVIDIA Expands Slurm So Supercomputing Centers Can Schedule Quantum Processors

    NVIDIA said the community-built Slurm workload manager now handles quantum processors, letting QPU makers expose their hardware to any supercomputing center through a single configuration file. Slurm 26.11 arrives later this month and builds on the existing QRMI and QDMI interfaces; Alice & Bob, IBM, IQM, Pasqal, Quantinuum and MQSC are already adopting it. Previously, each center-and-QPU pairing required its own custom software work.

  12. ProductsNVIDIA Newsroom
    NVIDIA's Vera Rubin Debuts in MLPerf With Leading Training and Agentic Results

    NVIDIA published the first MLPerf results for its Vera Rubin platform on Nov. 16, saying the system led every MLPerf Training v6.1 benchmark and topped its agentic tests in a first MLPerf Endpoints submission. NVIDIA reports Vera Rubin NVL72 delivered up to twice the performance per GPU of its GB300 NVL72 system. In a provisional Kimi K3 endpoint result, Vera Rubin showed more than 10x the throughput per watt of GB300 NVL72 at the same response speed; those figures await MLCommons verification. Twenty organizations submitted training results on NVIDIA platforms.

  13. ProductsNVIDIA Newsroom
    NVIDIA expands CUDA-X libraries to speed scientific research with AI agents

    NVIDIA introduced new and expanded CUDA-X libraries for scientific computing at SC26. cuPhoton, which includes the xBeam and xLens modules, correlated Event Horizon Telescope observations of the black hole jets J1924 and 3C279, reducing an analysis that once took over a month to eight minutes; the pipeline runs on multi-GPU systems or a single DGX Spark. cuPotential accelerates existing molecular dynamics software by up to 3x and is being adopted by drug-discovery companies including Cadence, Fathom Therapeutics and Terray Therapeutics. Daviator, arriving in NVIDIA ALCHEMI, speeds quantum-mechanical materials calculations. cuSpa is now open source for spatial biology, and the rapids-singlecell framework ships in an agent-ready container. NVIDIA said researchers can pair these pipelines with a locally running Nemotron-powered agent.

  14. ProductsNVIDIA Newsroom
    NVIDIA Announces Rubin HPC GPU for Scientific Simulation

    NVIDIA introduced Rubin HPC, a new GPU in its Vera Rubin platform, at the SC26 conference. It also uses seventh-generation NVLink for GPU-to-GPU connectivity. NVIDIA describes the design as balanced, scaling data movement alongside compute so recurring simulation patterns — sparse matrix operations, stencil computations, sorting and reductions — run faster. The company says Rubin HPC is expected to be available in 2028, and points to its CUDA-X libraries as optimized routines agents can call.

  15. ModelsNVIDIA Newsroom
    NVIDIA Releases Nemotron for Science With Open Chemistry Dataset

    NVIDIA introduced Nemotron for Science, a package that pairs its Nemotron 3.5 Super VL open model with a chemistry dataset and post-training recipes so researchers can build agents for scientific work. The model and dataset are available now on Hugging Face, and the model was optimized to run locally on a 128GB DGX Spark system. NVIDIA says Argonne National Laboratory, the Jülich Supercomputing Centre and companies including Terray Therapeutics and Lila Sciences are using or customizing the model for tasks such as extracting reactions from diagrams and designing laboratory protocols.

  16. ProductsNVIDIA Newsroom
    NVIDIA Expands Vera CPU Platform for Scientific Computing

    NVIDIA announced on November 16, 2026 a rack-scale platform built on its Vera data center CPU for high-performance computing. Each Vera chip carries 88 custom Olympus cores, 176 threads and SOCAMM2 LPDDR5X memory reaching up to 1.2TB/s. A rack connects 192 Vera CPUs over Quantum-X800 InfiniBand with roughly 300TB of memory and a petabyte of storage in a liquid-cooled MGX design. NVIDIA reports 1.9x the performance and 2.4x the performance per watt of x86 systems on an OpenFOAM airflow simulation, and 1.6x performance with 2x energy efficiency on the Weather Research and Forecasting model. Racks are expected from Dell, HPE, Lenovo and Supermicro in early 2027. HPC SDK 26.11 is available now, and a next-generation Fortran compiler is in public beta. DKRZ selected Vera for its HLRE-5 supercomputer, expected in early 2028.

  17. ProductsNVIDIA Newsroom
    NVIDIA, IBM and Fermilab Team Up on Quantum Codesign Platform

    NVIDIA said it is collaborating with IBM and the U.S. Department of Energy's Fermi National Accelerator Laboratory on a modular quantum computing codesign platform. NVIDIA DGX B200 systems are being colocated with IBM quantum computers, including systems powered by IBM Quantum Nighthawk, so researchers can develop and evaluate error-correction codes, decoders and quantum-classical applications for high-energy and nuclear physics. NVIDIA is also extending the open-source Slurm workload manager with a quantum interface, with IBM contributing integration through its Quantum Resource Management Interface.

  18. ProductsAWS Machine Learning
    Amazon Bedrock knowledge bases add tabular search for CSV and Excel files

    AWS added tabular search to Amazon Bedrock Managed Knowledge Base. When a question needs values or calculations from CSV, XLSX, or XLS files, agentic retrieval now copies the original file into an isolated sandbox with no network access and runs model-written code over the whole table or sheet, then combines the printed results with passages from related documents. It runs through the existing AgenticRetrieveStream request, so there is nothing to enable, and it works with files brought in through S3, SharePoint, Confluence and other connectors. Answers include citations, and the trace names the workbook the code read. Answers that use table data take longer, since running code adds at least one agent turn.

  19. ProductsNVIDIA Developer
    NVIDIA Vera Rubin NVL72 Debuts in MLPerf Training With Up to 2x Speedup

    NVIDIA's Vera Rubin NVL72 platform made its first MLPerf Training appearance in v6.1, posting the fastest time to train on six of the seven benchmarks and up to 2x more performance per GPU than an optimized GB300 NVL72 baseline at the same GPU count. NVIDIA says Vera Rubin NVL72 systems are generally available through cloud infrastructure partners.

  20. ProductsNVIDIA Developer
    NVIDIA Vera Rubin NVL72 Tops MLPerf Endpoints Agentic Inference Results

    NVIDIA published provisional MLPerf Endpoints v1.0 results for its Vera Rubin NVL72 rack-scale system, setting it against verified GB300 NVL72 figures across the Kimi K3, DeepSeek-V4.1-Flash and Qwen3.6-35B-A3B models. At a shared minimum of 50 tokens per second per user of end-to-end interactivity, the company reports more than 10x the throughput per provisioned kilowatt, with 5x to 10x gains for individual models and 3x to 5x higher interactivity at matched concurrency. The Rubin results are provisional and still pending MLCommons verification. Rubin GPUs pair HBM4 memory, Tensor Cores and sixth-generation NVLink; NVIDIA also credits prefix caching, state reuse and Dynamo routing for the serving gains.

  21. ProductsAWS Machine Learning
    Amazon Bedrock adds retrieval instructions for per-tenant answers

    AWS released retrieval instructions for Amazon Bedrock Managed Knowledge Base, a way to set how agentic retrieval answers for each tenant a request comes from. An instruction carries four fields: persona, answer format, retrieval guidance and answer policy, covering language, date and money conventions, citation style and whether to answer at all. Instructions belong to an AWS account and Region rather than to a knowledge base, go through an asynchronous review, and can be saved as numbered versions that applications pin so a change ships deliberately. A request names an instruction and optionally a version, and the planning trace records which was applied. AWS states that instructions shape voice, format and policy only; metadata filters or separate knowledge bases are still required to control what a tenant can retrieve.

  22. ProductsDatabricks
    Databricks launches AI Index tracking enterprise model adoption and costs

    Databricks has published the Databricks AI Index, a public resource updated weekly that tracks enterprise model adoption, AI spending patterns and coding-agent performance. The first edition reports that open-weight models grew from 7.3% of enterprise token traffic in July to 30.9% in November, and that the share of enterprises relying on a single model provider fell from roughly 64% to 28%. Databricks is also making its internal evaluation suite permanently public.

  23. ProductsVercel
    Vercel adds Tencent's Hy4 Flash to AI Gateway with 50% off intro pricing

    Vercel's AI Gateway now carries Hy4 Flash, Tencent's cheaper entry in the Hy4 model family, with an introductory 50% discount applied to input, cached input and output tokens through November 29, 2026. Vercel says the model keeps the family's full feature set — reasoning that preserves thinking across turns, tool calling, structured outputs, prompt caching and image input — inside the same 1M-token context window, at about a tenth of the per-token price. Reasoning can be switched off per request. Developers call it as tencent/hy4-flash, and can also pick it inside Claude Code, Codex and Cursor after installing the latest Vercel CLI and running setup.

  24. ProductsVercel
    Vercel adds Private Inference to AI Gateway for custom model endpoints

    Vercel's AI Gateway now supports Private Inference, which lets teams connect their own model endpoints — fine-tuned, self-hosted, or prerelease models — and run them through the same authentication, SDKs, APIs, and observability as catalog models. Requests can be made through Chat Completions, Responses, or Messages, with the Gateway translating each one to the endpoint's format so switching models only changes the model ID.