Tuesday, Nov 17

92 items

Showing the Nov 17 brief. Back to today ›

1 new rumor today →
Top stories
1OfficialProduct

OpenAI expands NATO cyber-defense work through Daybreak

OpenAI published this on Nov 17, 2026 (source: openai.com).

OpenAI says it is working with NATO's Communications and Information Agency to give NATO's core institutions access to advanced cybersecurity models through Daybreak. NATO has used up to 300 billion raw input and output tokens a month, and OpenAI also supported a 2026 Crossed Swords training stream in which participants built and evaluated AI agents with human oversight. The company says it has also provided cyber-model access to ENISA and national agencies in France, Germany and Poland.

Source: OpenAI
2OfficialProduct

OpenCode v1.19.6 lets OpenCode 1 upgrade to OpenCode 2.0

GitHub published this on Nov 17, 2026 (source: github.com).

OpenCode v1.19.6 lets the OpenCode 1 desktop app offer an upgrade to OpenCode 2.0. The app downloads OpenCode 2 in the background and installs it only when the user chooses Restart to upgrade, Upgrade on quit, or Later, and keeps existing sessions, projects, and settings.

Source: GitHub
Latest
Officialllama.cpp b12465 uses fast division in SYCL bin-bcast kernels

GitHub published this on Nov 17, 2026 (source: github.com).

llama.cpp prerelease b12465 replaces per-element modulo with fast_div_modulo in the SYCL bin-bcast kernels (#32213).

Source: GitHub
Officialllama.cpp b12464 adds an OpenCL A7 Q8_0 non-MoE binary kernel

GitHub published this on Nov 17, 2026 (source: github.com).

llama.cpp prerelease b12464 adds an OpenCL A7 Q8_0 non-MoE binary kernel (#31003).

Source: GitHub
Officialllama.cpp b12463 fixes Muse Glimmer vision embedding order

GitHub published this on Nov 17, 2026 (source: github.com).

llama.cpp prerelease b12463 fixes Muse Glimmer vision embedding order in mtmd (#32433).

Source: GitHub
Officialllama.cpp b12458 adds mtmd support for YuE2-3B

GitHub published this on Nov 17, 2026 (source: github.com).

llama.cpp prerelease b12458 adds mtmd support for YuE2-3B (#32369).

Source: GitHub
Officialllama.cpp b12457 adds an OpenCL A7 Q4_K non-MoE binary kernel

GitHub published this on Nov 17, 2026 (source: github.com).

llama.cpp prerelease b12457 adds an OpenCL A7 Q4_K non-MoE binary kernel (#30948).

Source: GitHub
Officialllama.cpp b12456 fixes OpenCL binary flash-attention prefill defects

GitHub published this on Nov 17, 2026 (source: github.com).

llama.cpp prerelease b12456 fixes three defects in the OpenCL binary flash-attention prefill path (#32110).

Source: GitHub
Officialllama.cpp b12455 implements an Inkling parser in chat

GitHub published this on Nov 17, 2026 (source: github.com).

llama.cpp prerelease b12455 implements an Inkling parser in the chat layer (#32381).

Source: GitHub
OfficialOpenAI publishes a Revolut story on ChatGPT, API, and Codex use

OpenAI published this on Nov 17, 2026 (source: openai.com).

OpenAI published a November 16 customer story on Revolut's use of ChatGPT, the API, and Codex in engineering and financial-crime work. Revolut says Codex has cut coding time by 30-60% from task start to feature merge, and that a FinCrime agent powered by OpenAI models can complete defined screening and investigation workflows after the release of GPT-5.5.

Source: OpenAI
OfficialElevenLabs schedules ElevenAgents API request timeouts for November 26

ElevenLabs published this on Nov 17, 2026 (source: elevenlabs.io).

ElevenLabs announced on November 16 that server-side request timeouts will apply to the ElevenAgents API (/v1/convai) starting November 26. During temporary latency increases, requests may return HTTP 504; clients should handle these responses, for example with backoff retries. This is a scheduled change, not a claim that enforcement has already begun.

Source: ElevenLabs
OfficialGitHub Copilot retires selected models with annual-plan and fallback exceptions

GitHub Changelog published this on Nov 17, 2026 (source: github.blog).

GitHub says Claude Opus 4.8, Claude Opus 5, and GPT-5.6 Terra were deprecated across most Copilot experiences on November 16. Opus 4.8 remains available to individual annual-plan subscribers, while Opus 5 remains available for runtime refusal fallback through existing authorized access despite leaving the model picker. GitHub recommends supported replacements; existing enterprise model policies still apply.

OfficialKiro Enterprise adds real-time telemetry and private connectivity

Kiro published this on Nov 17, 2026 (source: kiro.dev).

Kiro Enterprise now supports real-time OpenTelemetry events for LLM requests and completed tool calls, with delivery over public connections, AWS PrivateLink, or Amazon VPC Lattice. The features are available for profiles in US East (N. Virginia) and Europe (Frankfurt). Exports include usage data and user identifiers, not prompts, responses, code, or files; private connections do not fall back to public delivery.

Source: Kiro
OfficialKiro Web adds an in-session terminal

Kiro published this on Nov 17, 2026 (source: kiro.dev).

Kiro Web now lets users run commands in a session workspace from a terminal beside or below the conversation. Users can run builds and tests, inspect output, and try commands without leaving the session.

Source: Kiro
OfficialDevin adds subscribed-session views and replaces Archive with Mark done

Cognition published this on Nov 17, 2026 (source: docs.devin.ai).

Cognition's November 16 Devin update adds a sidebar view of subscribed sessions. Mark done replaces Archive and moves a session to the user's Done folder without stopping it or removing it from other subscribers' sidebars. Marking a session done does not mean its work has finished.

Source: Cognition
Officialllama.cpp b12470 fixes a SYCL flash-attention MKL mask-broadcast issue

GitHub published this on Nov 17, 2026 (source: github.com).

llama.cpp prerelease b12470 fixes a SYCL flash-attention MKL error when a mask is broadcast over heads or batches (#31536).

Source: GitHub
Officialllama.cpp b12469 adds a Metal implicit GEMM kernel for CONV_2D

GitHub published this on Nov 17, 2026 (source: github.com).

llama.cpp prerelease b12469 adds an implicit GEMM kernel for CONV_2D on Metal (#28069).

Source: GitHub
Officialllama.cpp b12468 implements timestep embedding on WebGPU

GitHub published this on Nov 17, 2026 (source: github.com).

llama.cpp prerelease b12468 implements timestep embedding on WebGPU (#31695).

Source: GitHub
Officialllama.cpp b12466 adds OpenCL SiLU support for f16 and non-contiguous f32

GitHub published this on Nov 17, 2026 (source: github.com).

llama.cpp prerelease b12466 adds OpenCL SiLU support for f16 tensors and non-contiguous f32 tensors (#32319).

Source: GitHub
OfficialNVIDIA publishes an experimental NVFP4-QAD checkpoint of Qwen3.8-Flash-Next

Hugging Face published this on Nov 17, 2026 (source: huggingface.co).

NVIDIA published Qwen3.8-Flash-Next-NVFP4-QAD on Hugging Face, an experimental mixed-precision QAD checkpoint of Qwen's Qwen3.8-Flash-Next exported after 100 optimizer updates with NVIDIA Model Optimizer and NeMo AutoModel. The card says the full multimodal model is included, with a frozen vision encoder, PLE n-gram embedding, and MTP module, and that the weight files occupy about 129.7 GB. NVIDIA says this recipe is more aggressive than its earlier Qwen3.8-Flash-Next-NVFP4 checkpoint, including NVFP4 shared experts and MXFP8 for selected projections, and that task-accuracy gains from QAD are not statistically established.

Source: Hugging Face
Officialllama.cpp b12477 rotates K/V read order for better L2 cache hit rates

GitHub published this on Nov 17, 2026 (source: github.com).

llama.cpp prerelease b12477 rotates K/V read order for better L2 cache hit rates (#32407).

Source: GitHub
Officialllama.cpp b12475 moves Llama Chat to SvelteKit 3 and its built-in service worker

GitHub published this on Nov 17, 2026 (source: github.com).

llama.cpp prerelease b12475 moves Llama Chat to SvelteKit 3 and its built-in service worker (#32463).

Source: GitHub
Officialllama.cpp b12474 applies an output-projection weight scale for Qwen3-Next

GitHub published this on Nov 17, 2026 (source: github.com).

llama.cpp prerelease b12474 applies an output-projection weight scale for Qwen3-Next (#32393).

Source: GitHub
Officialllama.cpp b12473 raises the CUDA shared-memory limit only for the launched kernel

GitHub published this on Nov 17, 2026 (source: github.com).

llama.cpp prerelease b12473 raises the CUDA shared-memory limit only for the launched kernel (#32392).

Source: GitHub
OfficialAWS publishes an ArisGlobal customer story on Bedrock prompt caching

AWS published this on Nov 17, 2026 (source: aws.amazon.com).

ArisGlobal runs NavaX, its life-sciences AI engine, on Amazon Bedrock. By caching the shared prompt prefix instead of each per-case document, ArisGlobal cut effective input cost by about 50% and overall monthly inference spend by about 40%, while onboarding more tenants without proportional cost growth.

Source: AWS
OfficialOpenAI and the Government of Chile launch OpenAI for Chile

OpenAI published this on Nov 17, 2026 (source: openai.com).

OpenAI launched OpenAI for Chile, a partnership with Chile's Ministry of Science, Technology, Knowledge and Innovation to support responsible AI use in public administration, train civil servants, and help local startups. The memorandum of understanding was signed on the sidelines of Semafor's Silicon Valley & The World conference. The MoU sets no investment or purchasing commitment.

Source: OpenAI
OfficialIsomorphic Labs raises $5 billion Series C to scale AI drug design

Isomorphic Labs published this on Nov 17, 2026 (source: isomorphiclabs.com).

Isomorphic Labs announced a $5 billion Series C funding round co-led by Greenoaks and Sequoia Capital, with participation from a16z, Alphabet and other investors. The company says the funding will support scaling its proprietary IsoDDE drug design engine and AI-designed therapeutic candidate pipeline. The announcement does not establish clinical efficacy or regulatory approval.

OfficialGitHub Copilot code review adds adaptive effort, secret detection, and cross-repository context

GitHub published this on Nov 17, 2026 (source: github.blog).

GitHub Copilot code review can now pick Lite, Balanced, or Max review depth from a pull request's complexity and risk, draw on related repositories, and flag exposed secrets before they reach the main branch. Adaptive, cross-repository context, and secret detection are in public preview for Copilot Pro, Pro+, Max, Business, and Enterprise plans. The same changelog also notes Copilot code review for Azure DevOps.

Source: GitHub
OfficialThe GitHub Copilot app adds Azure DevOps and GitHub Enterprise Server support

GitHub published this on Nov 17, 2026 (source: github.blog).

The GitHub Copilot app can now work with Azure DevOps repositories and a GitHub Enterprise Server instance, including cloning a repo, opening a pull request, and tracking status in the app. Azure DevOps integration is in public preview and is enabled from Settings > Experimental; cloud sessions are not available for Azure DevOps repositories. The same changelog also points to browser use in a Chromium-based browser with the Copilot Chrome extension.

Source: GitHub
OfficialGitHub Copilot benchmarks add Terminal-Bench-Science 0.1 and a harness comparison view

GitHub published this on Nov 17, 2026 (source: github.blog).

The GitHub Copilot benchmarks site, in public preview, now includes Terminal-Bench-Science 0.1 for research workflows across scientific domains, and a Harnesses view that compares GitHub Copilot CLI with Claude Code and Codex CLI using the same model at the same reasoning effort. The site already compared models in the Copilot agent harness on software-engineering benchmarks by resolution rate, reasoning effort, and cost per task.

Source: GitHub
OfficialGitHub Copilot CLI adds Microsoft Entra sign-in as a public preview

GitHub published this on Nov 17, 2026 (source: github.blog).

GitHub Copilot CLI now supports Microsoft Entra authentication in public preview on Windows and macOS, with Linux support to follow. Users can sign in with a Microsoft Entra account to authenticate to a linked GitHub Enterprise Managed User account, use a Microsoft 365 Copilot license, or both, depending on the organization's entitlements. EMU silent SSO requires an assigned Copilot Business or Enterprise license and an enabled Copilot CLI policy.

Source: GitHub
OfficialGitHub Copilot Evals is now in public preview

GitHub published this on Nov 17, 2026 (source: github.blog).

GitHub opened a public preview of Copilot Evals, a repeatable test for how a Copilot skill, custom agent, or MCP tool behaves after prompts, models, or tools change. Users can create an eval in their own words or with the /eval command, inspect passed and failed checks, and rerun the same eval to catch regressions. Each run must be approved before it starts.

Source: GitHub
OfficialGitHub Copilot Fusion expands to more surfaces and public Terminal-Bench leaderboards

GitHub published this on Nov 17, 2026 (source: github.blog).

GitHub said Copilot Fusion is now available on web, mobile, and Visual Studio, and is listed on Terminal-Bench 4.1 public leaderboards hosted by Stanford, Harbor, and the Laude Institute. The changelog also says Fusion gained additional models and appears on two public Terminal-Bench leaderboards, giving more ways to use Fusion and compare its performance.

Source: GitHub
OfficialGitHub Copilot browser use enters public preview in CLI, the app, VS Code, and JetBrains

GitHub published this on Nov 17, 2026 (source: github.blog).

Paired with the GitHub Copilot Chrome extension, browser use is now in public preview in GitHub Copilot CLI, the GitHub Copilot app, VS Code, and JetBrains. GitHub says users can ask Copilot to control the browser, debug web apps, automate browser tests, and handle complex web tasks. The extension settings can manage permission levels and experimental features.

Source: GitHub
OfficialGitHub Copilot adds MAI Code 2 and MAI Code 2 Flash

GitHub published this on Nov 17, 2026 (source: github.blog).

GitHub added Microsoft's MAI Code 2 and MAI Code 2 Flash models in GitHub Copilot for agentic coding workflows. Both models are billed under usage-based billing and are available to eligible subscribers on annual plans. Copilot Enterprise and Business administrators can manage access through the model policy in Copilot settings.

Source: GitHub
OfficialEnterprises can stream GitHub Copilot agent activity into Microsoft Agent 365

GitHub published this on Nov 17, 2026 (source: github.blog).

GitHub Enterprise Cloud customers with Enterprise Managed Users can now stream GitHub Copilot agent activity into Microsoft Agent 365, bringing audit data from Copilot clients into Microsoft's agent control plane to review alongside other AI workloads. GitHub points administrators to its documentation on streaming the enterprise audit log.

Source: GitHub
OfficialGitHub Copilot security review can run secret scanning on demand

GitHub published this on Nov 17, 2026 (source: github.blog).

Copilot's security review specialist can now run GitHub secret scanning alongside its existing LLM-based secret checks, in public preview in Copilot CLI and Copilot app sessions, Copilot SDK integrations, and Copilot code review. The review checks current changes and does not search repository history. Secret scanning is off by default; scans consume AI credits under the Secret Protection AI Credit SKU, and no additional plan is required if you can already use Copilot security review or code review.

Source: GitHub
OfficialGitHub adds Copilot code review outcome metrics in repository insights

GitHub published this on Nov 17, 2026 (source: github.blog).

A new Repository insights page, the first item in a repository's Insights tab, shows how Copilot code review feedback lands, including comments addressed, high-severity comments addressed, pull requests with no later human change request, and merged pull requests without Copilot review. The same metrics are also available in the enterprise and organization repos-1-day reports in the Copilot usage metrics API.

Source: GitHub
OfficialGitHub will retire MAI-Code-1.1-Flash in Copilot on December 18, 2026

GitHub published this on Nov 17, 2026 (source: github.blog).

With the launch of MAI Code 2 Flash, GitHub will deprecate MAI-Code-1.1-Flash across all GitHub Copilot experiences on December 18, 2026, and asks users to move workflows to the latest MAI Code Flash model before that date. Copilot Enterprise administrators may need to enable the replacement model in Copilot model policies.

Source: GitHub
Officialllama.cpp b12483 keeps the k-pool layout prefix on sequence edits

GitHub published this on Nov 17, 2026 (source: github.com).

llama.cpp prerelease b12483 keeps the k-pool layout prefix on sequence edits.

Source: GitHub
Officialllama.cpp b12482 speeds up the PLaMo-2/3 tokenizer table build

GitHub published this on Nov 17, 2026 (source: github.com).

llama.cpp prerelease b12482 speeds up the PLaMo-2/3 tokenizer table build.

Source: GitHub
Officialllama.cpp b12481 fixes FP8 Qwen4exp checkpoint conversion

GitHub published this on Nov 17, 2026 (source: github.com).

llama.cpp prerelease b12481 fixes FP8 Qwen4exp checkpoint conversion.

Source: GitHub
Officialllama.cpp b12480 adds a Llama UI design system for tokens, surfaces, and fonts

GitHub published this on Nov 17, 2026 (source: github.com).

llama.cpp prerelease b12480 adds a Llama UI design system for tokens, surfaces, and fonts.

Source: GitHub
Officialllama.cpp b12479 adds Minimax-Music3 support in mtmd

GitHub published this on Nov 17, 2026 (source: github.com).

llama.cpp prerelease b12479 adds Minimax-Music3 support in mtmd.

Source: GitHub
Officialllama.cpp b12478 adds a tts_dict for the Kokoro G2P template

GitHub published this on Nov 17, 2026 (source: github.com).

llama.cpp prerelease b12478 adds a tts_dict for the Kokoro G2P template in chat.

Source: GitHub
OfficialAWS describes secure OAuth patterns for Snowflake Cortex AI through Amazon Bedrock AgentCore

AWS published this on Nov 17, 2026 (source: aws.amazon.com).

AWS published a technical how-to on secure OAuth patterns for connecting Snowflake Cortex AI through Amazon Bedrock AgentCore, dated November 17, 2026.

Source: AWS
OfficialAWS publishes a how-to on building WhatsApp AI agents with Amazon Bedrock AgentCore

AWS published this on Nov 17, 2026 (source: aws.amazon.com).

AWS published a technical how-to on building WhatsApp AI agents on Amazon Bedrock AgentCore, dated November 17, 2026. The post focuses on integration work such as authenticating and normalizing inbound WhatsApp payloads, processing media, conversation memory, and Meta message templates for outbound messages sent more than 24 hours after the user's last message.

Source: AWS
OfficialAWS publishes a LoanPro customer story on an agentic loan-servicing assistant

AWS published this on Nov 17, 2026 (source: aws.amazon.com).

AWS, with LoanPro, published a customer story on how LoanPro built an agentic AI loan-servicing assistant with Amazon Bedrock AgentCore, dated November 17, 2026. The post says the assistant is meant to pull account data so servicing agents spend less time assembling context by hand during borrower calls.

Source: AWS
OfficialMicrosoft makes Copilot Home and Office generally available and starts Autopilot for Frontier customers

Microsoft published this on Nov 17, 2026 (source: microsoft.com).

At Microsoft Ignite 2026, Microsoft said Copilot Home, where Chat and Cowork come together, and Office in Copilot are now generally available, and that Copilot Managed Runtime for hosting and managing apps is also generally available. Autopilot is beginning to roll out to Frontier customers, Code will become generally available in the coming weeks, and new plugins and skills for sales, service, finance, and legal teams are available.

Source: Microsoft
OfficialOpenAI introduces GPT-6.2 Astra and GPT-6.2 Sol

OpenAI published this on Nov 17, 2026 (source: openai.com).

Starting today, Plus, Pro, Enterprise, and Edu users can use GPT-6.2 Astra and GPT-6.2 Sol in ChatGPT Work and Codex, and developers can call gpt-6.2-astra and gpt-6.2-sol in the API, with Ultrafast available for both models at launch. OpenAI describes Astra as its new frontier model for harder professional work and Sol as the efficient everyday model. Astra API input and output prices are 20% lower than GPT-6 Astra ($8 / $0.40 cached / $40 per million tokens); Sol keeps GPT-6.1 Sol prices. Eligible Enterprise and Edu users get access once a workspace admin enables it.

Source: OpenAI
OfficialOpenAI updates Astra for Law on GPT-6.2 Astra and opens Enterprise access requests

OpenAI published this on Nov 17, 2026 (source: openai.com).

OpenAI released an updated Astra for Law configuration of GPT-6.2 Astra, with an expanded legal search index, and said ChatGPT Enterprise customers can now request access through the Trusted Access Program, including through the API. OpenAI reports the configuration ranked first on Harvey LAB-AA, earning full credit on 20.8% of assignments, ahead of GPT-6.2 Astra at 15.0%. The post does not claim legal or clinical efficacy.

Source: OpenAI
OfficialGoogle makes Tasks in Gemini free in six countries and adds new U.S. features

Google published this on Nov 17, 2026 (source: blog.google).

Google said Tasks in Gemini is now free of charge in six countries, with new features starting in the U.S. and deeper Google Workspace integration so Gemini can get things done on a user's behalf. The official post is dated November 17, 2026.

Source: Google
OfficialOpenCode v2.3.1 keeps timed-out shell commands running and writes compaction in the background

GitHub published this on Nov 17, 2026 (source: github.com).

OpenCode v2.3.1, published on November 17, moves a shell command that is still running when its timeout expires to the background instead of killing it, and writes the session-compaction summary in the background so the next prompt does not wait. A foreground timeout of 0 now means the default 120 seconds. The release also adds GPT-6.2 Sol and GPT-6.2 Astra, including fast variants, to the model picker with ChatGPT sign-in, lets users update plugins from Settings, and adds an Add comment toolbar on selected code in diff and file viewers.

Source: GitHub
Officialllama.cpp b12487 adds a CUDA few-row MMQ path for MUL_MAT_ID

GitHub published this on Nov 17, 2026 (source: github.com).

llama.cpp b12487, published on November 17, adds a CUDA few-row MMQ path for MUL_MAT_ID, including a shared launcher and a earlier threshold on DGX Spark.

Source: GitHub
Officialllama.cpp b12485 uses CUDA block reductions for whole-row pool_1d

GitHub published this on Nov 17, 2026 (source: github.com).

llama.cpp b12485, published on November 17, switches whole-row pool_1d work to CUDA block reductions and templates the pool1d row kernel by ggml_op_pool.

Source: GitHub
Officialllama.cpp b12484 decodes F8_E4M3 to fp32 without ldexpf

GitHub published this on Nov 17, 2026 (source: github.com).

llama.cpp b12484, published on November 17, changes ggml so F8_E4M3 values decode to fp32 without calling ldexpf.

Source: GitHub
OfficialAWS describes ways to cut token costs and latency in Bedrock AgentCore JSON workloads

AWS published this on Nov 17, 2026 (source: aws.amazon.com).

AWS published a technical how-to on reducing token costs and latency when production agents on Amazon Bedrock AgentCore process multi-megabyte JSON payloads, dated November 17, 2026. The post describes three composable strategies, including offloading heavy artifacts.

Source: AWS
OfficialOpenAI introduces ChatGPT for Sales with a Sales plugin and expanded partner plugins

OpenAI published this on Nov 17, 2026 (source: openai.com).

OpenAI introduced ChatGPT for Sales, with a new Sales plugin and more than 30 new and expanded partner plugins in ChatGPT to research prospects, prepare pitches, update a CRM, and move deals forward. The official post is dated November 17, 2026.

Source: OpenAI
OfficialPerplexity introduces Orchestrator API for model and provider routing

@perplexity_ai on X published this on Nov 17, 2026 (source: x.com).

Perplexity introduced Orchestrator API, combining open-weight and proprietary models behind one API and unified billing. Users choose a model for a task, while the service selects a provider based on availability, capacity and latency.

Part of Perplexity
OfficialReplit introduces Missions for agent teams working within a budget

@Replit on X published this on Nov 17, 2026 (source: x.com).

Replit introduced Missions, an agent-swarm harness designed to pursue objectives within user-defined boundaries and budgets. Replit describes it as a way to coordinate autonomous work; its DataCat example is part of the same launch, not another product.

Part of Replit
Source: @Replit on X
OfficialGitHub and Microsoft describe MDASH security-report verification

@github on X published this on Nov 17, 2026 (source: x.com).

GitHub said its Codename MDASH work with Microsoft Security reduced 9,791 raw AI security findings to 99 verified, private reports with suggested fixes for six open-source projects. GitHub leadership said 55 issues had been fixed. These are the companies' reported results, not an independent audit.

Part of Microsoft
Source: @github on X
OfficialMicrosoft introduces an AI Transparency Initiative

@Microsoft on X published this on Nov 17, 2026 (source: x.com).

Microsoft introduced an AI Transparency Initiative at Ignite, saying it will work across the ecosystem on open standards for AI transparency, security and governance, including independently verifiable systems.

Part of Microsoft
OfficialMicrosoft Perception becomes generally available

@Microsoft on X published this on Nov 17, 2026 (source: x.com).

Microsoft announced general availability of Perception, which brings red-, blue- and green-team agents together through playbooks for attack simulation, threat detection and remediation.

Part of Microsoft
OfficialMicrosoft presents Agent 365 Business Analytics for AI spending and outcomes

@Microsoft on X published this on Nov 17, 2026 (source: x.com).

Microsoft presented Agent 365 Business Analytics at Ignite as a way to identify what drives AI spending, connect costs to business outcomes and compare value across use cases.

Part of Microsoft
OfficialGitHub details the open Agent Host Protocol for cross-device agent sessions

GitHub published this on Nov 17, 2026 (source: github.blog).

In a November 17 engineering post, GitHub explained how the open Agent Host Protocol keeps agent sessions synchronized across VS Code, the GitHub Copilot app and cloud environments, and enables custom clients and integrations. The post describes libraries in Rust, TypeScript, Kotlin, Swift, .NET and Go.

Source: GitHub
Officialeve 1.0 becomes generally available for building production agents

@eve on X published this on Nov 17, 2026 (source: x.com).

eve announced general availability of version 1.0, with support for using a developer's own harness, multi-tenant and multiplayer agents, session governance and tracing through Agent Runs, and agent interfaces in Next.js, Nuxt and other frameworks. Vercel shared the announcement.

Part of Vercel
Source: @eve on X
OfficialMeta says three Muse models are coming to Microsoft Foundry

@finkd on X published this on Nov 17, 2026 (source: x.com).

Mark Zuckerberg said Muse Wonder, Muse Spark 1.3 and Muse Ember are coming soon to Microsoft Foundry on Azure. No general-availability date or pricing was given in the post.

Part of Meta Muse
Source: @finkd on X
OfficialMicrosoft introduces IQ for shared organizational context and agent playbooks

@Microsoft on X published this on Nov 17, 2026 (source: x.com).

Microsoft presented IQ at Ignite as a way to bring business context together so agents can use organizational knowledge through shared plugins and playbooks.

Part of Microsoft
OfficialMicrosoft introduces Copilot Workspaces for people and agents

@Microsoft on X published this on Nov 17, 2026 (source: x.com).

Microsoft introduced Copilot Workspaces, describing a multiplayer environment in which people and agents collaborate toward a common goal with shared context.

Part of Microsoft
OfficialGoogle AI Studio improves Android emulation, data support and publishing

@GoogleAIStudio on X published this on Nov 17, 2026 (source: x.com).

Google AI Studio updated its Android app-building experience with HD emulator resolution and audio, Firestore database and authentication support, and publishing to a personal Play library. Personal Play-library publishing is currently limited to the United States.

OfficialMidjourney adds Moodboards and style references to its edit model

@midjourney on X published this on Nov 17, 2026 (source: x.com).

Midjourney added support for Moodboards, Style Codes and Style References in its edit model. Users can attach an image and add the style inputs without supplying a text prompt.

Part of Midjourney
OfficialCognition introduces Devin Bots to coordinate teams of agents

@cognition on X published this on Nov 17, 2026 (source: x.com).

Cognition introduced Devin Bots, describing a lead bot that coordinates a team of agents on software work. The company cited firmware development and large volumes of pull requests as examples, not guarantees of performance for every user.

Part of Devin
OfficialGoogle makes personal avatars in Vids free for everyone on desktop

@GoogleWorkspace on X published this on Nov 17, 2026 (source: x.com).

Google Workspace said personal avatars in Google Vids are now available at no cost to everyone, with the feature accessed on desktop.

Part of Google
OfficialRunway Labs previews Diffusion-UI research

@runwayml_labs on X published this on Nov 17, 2026 (source: x.com).

Runway Labs shared an early look at Diffusion-UI, a research application in which a real-time model generates interactive interfaces. The team says the interfaces can use structured data, maintain state and generate controls.

Part of Runway
OfficialPerplexity rolls out Multiagent in Computer and teammates

@perplexity_ai on X published this on Nov 17, 2026 (source: x.com).

Perplexity is rolling out Multiagent as a new effort level in Computer and teammates. A lead agent delegates work to agents using different frontier models, which share findings before the lead agent combines the result.

Part of Perplexity
OfficialSuno introduces Visuals for clips and lyric videos

@suno on X published this on Nov 17, 2026 (source: x.com).

Suno introduced Visuals for creating clips, lyric videos and other visuals around music, and confirmed that Visuals is included in its MCP integration. It offered paid subscribers one free first Visual for a limited time.

Part of Suno Studio
Source: @suno on X
OfficialKling introduces Image 4.0 with native 8K generation

@Kling_ai on X published this on Nov 17, 2026 (source: x.com).

Kling introduced Image 4.0 and highlighted native 8K image generation. Its teaser and the later resolution demonstration describe the same model launch.

Part of Kling
OfficialGemini side panel in Google Apps Script becomes generally available

@GoogleWorkspace on X published this on Nov 17, 2026 (source: x.com).

Google Workspace announced general availability of the Gemini side panel in Apps Script. It can generate scripts from plain-language requests, preview proposed code changes block by block and help fix runtime errors.

Part of Google
OfficialNVIDIA board establishes a national-security and export-control oversight committee

@nvidianewsroom on X published this on Nov 17, 2026 (source: x.com).

NVIDIA announced that its board has established a National Security and Export Control Oversight Committee. The announcement is a board-governance development, not a statement about the outcome of any investigation.

Part of NVIDIA
OfficialQwen4-Max becomes available on LebrelAI

@Alibaba_Qwen on X published this on Nov 17, 2026 (source: x.com).

Qwen announced that Qwen4-Max is now available on LebrelAI for agentic coding. This is a platform-integration announcement, not a new Qwen4-Max model release.

Part of Qwen
OfficialQwen4-Max becomes available on You.com

@alibaba_cloud on X published this on Nov 17, 2026 (source: x.com).

Alibaba Cloud and Qwen announced that Qwen4-Max is now available on You.com, adding another way to access the existing model.

Part of Qwen
OfficialLolipop AI Gateway adds Qwen4-Max and DeepSeek-V4.1-Flash-1017 via Alibaba Cloud

@alibaba_cloud on X published this on Nov 17, 2026 (source: x.com).

Alibaba Cloud said Qwen4-Max and DeepSeek-V4.1-Flash-1017 are now accessible through Lolipop AI Gateway via Alibaba Cloud Model Studio.

OfficialGemini Notebook adds source-based interactive sites

@Google on X published this on Nov 17, 2026 (source: x.com).

Google said Gemini Notebook can now turn a user's sources into interactive sites, including dashboards and visual explainers. Users describe the site in chat and can open it in a new tab to view or share it.

Source: @Google on X
OfficialGoogle DeepMind adds an Antigravity robotics plugin for Gemini 4 Hydrogen

@GoogleDeepMind on X published this on Nov 17, 2026 (source: x.com).

Google DeepMind described a new robotics plugin for Antigravity with Gemini 4 Hydrogen. Users can connect a supported robotic arm, practice tasks in MuJoCo simulation and run tasks on real hardware through conversation.

OfficialMeta launches Muse in Australia

@alexandr_wang on X published this on Nov 17, 2026 (source: x.com).

Alexandr Wang announced that Meta's Muse is now available in Australia. The announcement is a regional expansion, not a new underlying model release.

Part of Meta Muse
OfficialElevenLabs adds Scribe v3 Live Medical to ElevenAgents

@ElevenLabs on X published this on Nov 17, 2026 (source: x.com).

ElevenLabs said Scribe v3 Live Medical is now in ElevenAgents for transcription in live calls. Teams can supply up to 10,000 custom terms, including provider and medication names.

Part of ElevenLabs
OfficialInfron lists Hy4 and its Hy4 Uncensored variant behind one API

@InfronAI on X published this on Nov 17, 2026 (source: x.com).

Infron said it has made tencent/hy4 and tencent/hy4-uncensored available through one API for comparison. The uncensored offering is the provider's claim, not an endorsement by Tencent or an independently verified safety assessment.

Part of Infron
OfficialVercel AI Gateway adds Private Inference for private models

@vercel on X published this on Nov 17, 2026 (source: x.com).

Vercel added Private Inference to AI Gateway. Developers can access their own fine-tuned, self-hosted or prerelease models alongside catalog models using the same APIs, authentication and logs, with a private/<slug> model name.

Part of Vercel
Source: @vercel on X
OfficialMicrosoft announces Code 2, Code 2 Flash and Cyber 2 Flash

@Microsoft on X published this on Nov 17, 2026 (source: x.com).

Microsoft announced Code 2 for repository-wide reasoning, Code 2 Flash for high-volume coding and Cyber 2 Flash for security work at Ignite. This model-family announcement is distinct from the already-published GitHub Copilot integration.

Part of MAI
OfficialMicrosoft says MAI Image 2.7 ranks first on Artificial Analysis image editing

Microsoft published this on Nov 17, 2026 (source: microsoft.ai).

Microsoft said at Ignite on November 17, 2026, that MAI Image 2.7 is No. 1 on the Artificial Analysis image-editing leaderboard, outperforming GPT Image 2.6 Sunburst (max) at one fifth of the price. Artificial Analysis lists MAI-Image-2.7 as leading AA-Image-Editing v2.1 with an Elo score of 1215. Microsoft also said MAI Image 2.7 and MAI Image 2.7 Flash are available in Microsoft Foundry. The announcement does not describe that Foundry availability as a public preview. This item covers the Image 2.7 ranking and Foundry availability, not the separately published MAI Code 2 and Cyber 2 announcements.

Part of MAI
Source: Microsoft
OfficialAnthropic adds Stable and Latest release channels to Claude Desktop

Anthropic published this on Nov 17, 2026 (source: support.claude.com).

Anthropic's November 17, 2026 release notes say Claude Desktop now lets users choose a release channel that sets which version they get when the app updates. Stable stays the default and gets new versions less often, after more testing. Latest gets a new version about once a day, so fixes and new features reach users sooner. Anthropic's official install guide, which the notes link to for details, says users change the channel under Settings > Desktop app, in the Release channel control, and that Latest can be less reliable than Stable. The guide also says the channels are available on Claude for Mac and Claude for Windows, and are not available on Linux, Microsoft Store installs, or computers where IT has set a Claude policy.

Source: Anthropic
OfficialApple releases FastFrames official implementation and model checkpoints

Apple published this on Nov 16, 2026 (source: huggingface.co).

Apple published official implementations and model checkpoints for FastFrames-S (1.3B parameters) and FastFrames-L (14B) on Hugging Face. The model card describes both as autoregressive, two-step video diffusion models. The official repository says generated outputs are 832 x 480 video, 81 frames at 16 fps. 16 fps is the output frame rate, not a real-time generation speed. Hugging Face first created the model repositories on November 16, 2026.

Part of Apple
Source: Apple