AI news for Nov 17, 2026
13 stories
- ModelsOpenAIOpenAI releases GPT-6.2 Astra and Sol
OpenAI introduced GPT-6.2 Astra and Sol, two new models the company says deliver better professional outputs, coding and computer use, with improved efficiency at lower cost. They are available across ChatGPT Work, Codex and the API, according to OpenAI's announcement.
- ProductsGoogleGoogle rolls out free task-running features in Gemini to six countries
Google said its Gemini app can now carry out multi-step chores for users — calling businesses, booking appointments, comparing products, checking out, and adding items to grocery orders — running on a new Gemini 4 Hydrogen model. The task features are rolling out free to users in the U.S., Japan, Korea, Canada, Australia and India, with a set of advanced features starting in the U.S. Gemini connects to Google Password Manager, Google Pay and more than 60 apps, and accepts requests forwarded to [email protected] or sent by text. Google also said Wallet is expanding to iOS and that developers can apply to its Connected Apps Platform.
- ResearchHacker NewsRunta and Baseten benchmark 22 coding agent harnesses across five models
Runta published FrontierHarness Eval v1.1, a harness leaderboard built with Baseten, expanding its earlier single-model version to 22 harness configurations run against five models: GLM-5.3, GLM-5.3 Flash, DeepSeek V4.1 Flash, Muse and Kimi-K3. The main finding is that harness rankings shift with the model: Codex CLI was cheapest per task with GLM-5.3 but 16th with Kimi-K3, and with a fixed model the priciest harness cost up to five times the cheapest one. Step count correlated with cost per pass; prompt size, tool count and cache hit rate did not.
- ProductsNVIDIA NewsroomNVIDIA Announces Nemotron for Science Toolkit and Rubin HPC GPU at SC26
NVIDIA used its SC26 special address in Chicago to announce Nemotron for Science, a toolkit that pairs the open Nemotron 3.5 Super VL model with an open chemistry dataset and post-training recipes for chemistry and life sciences. Ian Buck, the company's VP of hyperscale and high-performance computing, said researchers can run it on a DGX Spark, a workstation or their own supercomputer. NVIDIA, IBM and Fermilab announced plans for a quantum co-design platform.
- ProductsAWS Machine LearningAWS adds AI data preparation tools to Amazon SageMaker
AWS announced AI data preparation capabilities in Amazon SageMaker AI, generally available in US East (N. Virginia) and US West (Oregon). The building blocks cover auto-labeling that uses agreement among several models as a confidence score, human review through SageMaker Studio built on Label Studio Community Edition, synthetic data generation grounded in seed documents, quality evaluation for duplicates and answer leakage, and transformation into formats such as SFT, DPO and RL. Users reach the capabilities through SageMaker Python SDK APIs or an AI-ML agent skill in notebooks and IDEs including Kiro. Source data, generated datasets and labels stay in the customer's own S3 buckets under an IAM role they control.
- BusinessAWS Machine LearningAWS expands Forward Deployed Engineering to 15 countries
AWS says its Forward Deployed Engineering service, introduced in June, is now offered in 15 countries — including the U.S., Canada, Mexico, Brazil, the UK, Germany, France, the Netherlands, Sweden, India, Japan, South Korea, Singapore, Australia and Saudi Arabia — for commercial and public sector customers. Small AWS teams of engineers, scientists and product managers work alongside customer staff in 45-day sprints inside the customer's own AWS account, with the goal of taking one high-value workload to production within a quarter. The offering includes prebuilt accelerators for tasks such as document processing and customer sentiment analysis, and runs on Amazon Bedrock AgentCore with governance and cost controls. AWS names Samsung Life Insurance, Rio Tinto Iron Ore and Marsh among early customers.
- ProductsWindowsMicrosoft adds agentic browsing to Edge for Business at Ignite 2026
At Microsoft Ignite 2026, Microsoft announced that Edge for Business now includes agentic browsing in Cowork, available through the sidepane, new tab page, and web. It requires a Microsoft 365 Copilot license with usage-based billing enabled. A skills recorder, coming in preview by year end, turns recorded workflows into reusable Cowork skills. Managed agent browsing, in preview with Microsoft 365 E5, lets IT run agents from Microsoft and other providers in a governed browser with Microsoft Entra agent identities. Microsoft also introduced Remix, generally available, for customizing websites without code, and an interactive brief feature that turns tab content into a shareable page.
- ProductsWindowsMicrosoft adds agent containment and local Copilot models at Ignite 2026
Microsoft outlined how Windows will host AI agents at its Ignite 2026 conference, in a post from Pavan Davuluri, executive vice president of Windows and Devices. New containment capabilities in Microsoft Execution Containers, which OpenAI and Meta are already building on, target running agents securely, and Windows 365 for Agents is now eligible for Copilot Credits. GitHub Copilot Fusion is in preview, letting developers route coding work to local models on their PCs, while local context, actions and models are promised for Copilot on Copilot+ PCs. Windows Server 2028 is in public preview with Microsoft Foundry and support for new AI hardware from AMD and NVIDIA.
- ProductsWindowsMicrosoft adds Windows security controls for AI agents and post-quantum TLS
Microsoft used its Ignite 2026 event to lay out Windows security changes aimed at AI-driven threats. The Windows endpoint security platform is now generally available on Windows 11, giving security vendors a path that reduces reliance on kernel drivers. Windows baseline protection, in preview, combines Secure Boot, virtualization-based security and memory integrity into a single versioned security state that users, IT and developers can check and verify through hardware-backed attestation. The Windows TLS stack is adding pure post-quantum key exchange and authentication. Windows Hello gets more consistent passwordless sign-in, IT observability into biometric and PIN use, and easier security key handling. Intune and Windows Autopatch are extending to eligible Windows Server scenarios in preview, and Microsoft Execution Containers provide governed boundaries for agent activity.
- ProductsAWS Machine LearningOpenTSLM time-series model is now available in Amazon SageMaker JumpStart
AWS announced that OpenTSLM, an open-source family of time-series language models from Aionic Labs, is now listed in Amazon SageMaker JumpStart, letting teams deploy it inside their own AWS environments. The available checkpoint, OpenTSLM-Flamingo, pairs a Gemma 3 4B language model with a time-series encoder and cross-attention layers, for 5.24 billion parameters total, and runs on a single GPU with at least 24 GB of memory. AWS also published a walkthrough that fine-tunes a Hugging Face checkpoint on ARFBench data loaded through TimeNet and serves the result on a SageMaker real-time endpoint. Nexthink described applying the model in Spark, its employee IT agent, to read device telemetry.
- ProductsSalesforceInformatica from Salesforce brings data governance into Microsoft Copilot and Foundry
Informatica, the Salesforce data management unit, announced at Microsoft Ignite 2026 an expanded collaboration with Microsoft that places its catalog, business glossary, lineage, data quality and governance signals inside Microsoft's AI products. A new Copilot and Teams integration lets employees ask in natural language what data means, where it came from and how trustworthy it is. In Microsoft Foundry, updated Informatica Headless Model Context Protocol servers add governance, identity and administrative controls for agents built on enterprise data. Fabric scanning now covers notebooks and data pipelines for broader lineage visibility, and Informatica added an Azure U.S. East Coast point of delivery for regional residency needs.
- BusinessSalesforceSalesforce and Microsoft Launch Copilotforce to Bring CRM Data Into Copilot
Salesforce and Microsoft announced Copilotforce at Microsoft Ignite, a partnership that puts Salesforce customer context and sales workflows inside Microsoft Copilot. Sellers are to prepare for meetings, review opportunities, flag deal risks and take supported CRM actions using context from Teams conversations, Outlook email and shared documents, with writebacks requiring explicit user confirmation and honoring existing Salesforce permissions and validation rules. A separate bidirectional zero-copy link lets organizations query supported Fabric OneLake data inside Salesforce Data 360 and share Data 360 context into OneLake without duplicating datasets. Sales Cloud in Teams, shown at Dreamforce, remains in pilot. Accenture is design and implementation partner and customer zero, deploying the integrations across its own sales organization.
- ProductsDatabricksDatabricks opens Azure Government beta with Unity Catalog and serverless
Databricks says Azure Databricks on Azure Government is now in Beta, built on the same architecture as the commercial service. The first release gives U.S. agencies and their partners Unity Catalog for access control, lineage and auditing, Databricks SQL with classic or serverless compute, workspaces with the Compliance Security Profile enabled by default, and system tables covering audit logs, billing and compute. Customer content stays in US Gov Virginia and US Gov Arizona. Databricks says AI capabilities — Genie, Agent Bricks, Model Serving, Vector Search, Lakebase — will follow, with FedRAMP Class D and DoW IL5 authorization expected in early 2027. Beta is intended for non-production workloads, and workspaces built now carry forward to general availability.