AI news daily brief: 2026-11-14
Seven stories from the past two days, led by a GitHub Copilot model picker that absorbed a third new entry in four days while GitHub locked in a hard cutoff for six old ones: two more Copilot platform updates, a shared-teammate upgrade from xAI, a two-week-only model preview from Alibaba, a scriptable security API from GitHub, and a Claude Code patch for a decoding bug.
Gemini 4 Hydrogen reaches general availability in GitHub Copilot
Google's Gemini 4 Hydrogen reached general availability in GitHub Copilot on November 13, the third new Gemini entry to land in Copilot's model picker in four days, following Gemini 4 Carbon on November 10 and the Claude Fable 5.2 default switch on November 11 and 12, per GitHub's changelog. GitHub describes Hydrogen as a Flash-class model built for fast agentic coding, tool use, and multistep workflows, with better instruction-following than the Gemini 3.8 Flash it is meant to replace. It rolled out across Copilot Pro, Pro+, Max, Business, and Enterprise, selectable in VS Code, Visual Studio, the CLI, JetBrains IDEs, Xcode, and every other major Copilot surface, and bills at usage-based provider rates like Copilot's other non-default models. New models enable by default for an org unless an admin has turned off the global setting, so Hydrogen can show up in a team's picker without anyone opting in. Full story: GitHub Copilot's model picker gets a third new entry in four days.
The GitHub Copilot app adds native GitHub Enterprise Server support
GitHub moved GitHub Enterprise Server (GHES) account support in the standalone Copilot app from opt-in preview to general availability on November 13, available in app version 1.2.0 and later, per GitHub's changelog. Self-hosted GHES customers sign in with a GitHub.com or GHE.com account first, then add their GHES account under Settings, Accounts, Add account, and can work with remote repositories, issues, and pull requests from inside the app. Cloud and background-agent sessions still are not available for GHES accounts, so a GHES user runs agent tasks locally rather than handing them to Copilot's hosted runners. Pondero folded the update into its existing Copilot Teams integration guide. Full story: GitHub Copilot in Microsoft Teams: What It Does, Who Gets It, and How to Turn It On.
GitHub sets December 14 as the deprecation date for six Copilot model options
GitHub set December 14, 2026 as the retirement date for six Copilot models across every surface, including Chat, inline edits, ask and agent modes, and code completions, per GitHub's changelog. The list: Claude Fable 5 (replaced by Fable 5.2), Claude Haiku 4.5 (individual subscribers on annual plans keep access), Gemini 3.8 Flash (replaced by today's GA'd Hydrogen), Gemini 4 Argon (replaced by Carbon), Grok 4.6 (replaced by Grok 4.7), and Kimi K3, whose successor GitHub has not named yet. Under default model enablement, the suggested replacements turn on automatically for Business and Enterprise accounts unless an admin disabled that global setting; teams that locked down their own model policy have to add the replacements by hand. GitHub says no action is needed to remove the deprecated models themselves once the date passes. Pondero folded the full list into its Copilot model-deprecation tracker. Full story: GitHub Copilot deprecated models: what to switch to and what it costs.
xAI's Grok Bot graduates to 1.0, becomes a shared teammate in Slack, Teams, and video calls
xAI moved Grok Bot out of beta on November 2 and turned it from one person's private assistant into a teammate a whole group can share, per xAI's announcement. "Multiplayer" Bots can be brought into Slack, Microsoft Teams, and video calls on Google Meet, Zoom, and Teams, where they track who is speaking and what has already been decided. "Proactive" Bots can, in xAI's words, "recognize work that needs to be done and offer to take it on" instead of waiting to be asked. Access rides on xAI's existing subscription tiers rather than a separate price tag: Grok Bot ships with the $30-per-month Premium plan, and a Teams plan starts at $30 per user per month. Enterprise admins still have to switch on Meetings access by hand. Full story: xAI turns Grok Bot into a shared AI teammate across Slack, Teams, and video calls.
Alibaba opens a two-week preview of Qwen4-Max with a 1-million-token context window
Alibaba's Qwen team opened a public preview of Qwen4-Max on November 11, running until November 25 (10:00 to 24:00, UTC+8), with a 1-million-token context window and a maximum output of 393,216 tokens shared between the model's reasoning and its final answer, per Alibaba Cloud's Model Studio documentation. The model takes text, image, and video input for tasks like document understanding and video analysis, and supports web search, tool calling, and an adjustable reasoning-effort parameter through either the OpenAI or Anthropic SDK format. None of it is reachable through a metered, pay-as-you-go key: a Token Plan subscription, sold in Personal or Team tiers, is the only way in during the preview. Alibaba published no benchmark table, parameter count, or license terms alongside the release, a gap for a team whose earlier Qwen launches often shipped open weights on day one. Full story: Alibaba opens a 14-day preview of Qwen4-Max with a 1-million-token context window.
GitHub lets admins manage AI Scan at the individual-repository level via the REST API
GitHub added a repository_scan setting to its AI Scan REST endpoints on November 13, letting admins turn the AI-powered vulnerability scanner on or off per repository instead of only through the web UI, per GitHub's changelog. The setting works through PATCH /orgs/{org}/code-scanning/ai-scan and a matching repo-level endpoint: an organization that enables it cascades to every repository, but individual repo admins can opt back out, and an org that disables it blocks repositories from turning it on themselves. Setting pr_scan and repository_scan together in one request means a bad bulk update fails as a whole with a 422 response rather than leaving some repositories half-configured. Each scan still consumes GitHub Code Security AI credits, so the same endpoint doubles as a scripted budget lever for security teams rolling AI Scan out across thousands of repositories at once. Full story: GitHub adds a per-repository API toggle for its AI code-vulnerability scanner.
Claude Code 2.1.325 patches a compressed-response decoding security bug and tightens permission prompts
Anthropic shipped Claude Code 2.1.325 on November 13 with 88 bug fixes in one release, including a fix for a security issue in how zstd-compressed responses are decoded, per the Claude Code changelog. The same release narrows several defaults that previously ran without asking: sandboxed Bash commands now ask first unless sandbox.autoAllowBashIfSandboxed is set to true, read-only git commands ask before running inside a repository that is not the project's own or one already trusted, and scheduled tasks and /loop now ask before running a skill that declares allowed-tools or hooks unless that skill is added to permissions.allow. None of the four changes adds a new capability. Each one narrows what Claude Code will do without a prompt first, shipped in the same release that closed a decoding-layer security bug Anthropic has not detailed further.
Sources
- Gemini 4 Hydrogen in GitHub Copilot: GitHub changelog, November 13, 2026
- GitHub Enterprise Server support in the GitHub Copilot app: GitHub changelog, November 13, 2026
- Upcoming deprecation of selected GitHub Copilot models in December: GitHub changelog, November 13, 2026
- Introducing Grok Bot 1.0: xAI announcement, November 2, 2026
- qwen4-max-preview - Alibaba Cloud Model Studio documentation: Alibaba Cloud, primary
- Manage AI Scan for repositories using the REST API: GitHub changelog, November 13, 2026
- Claude Code changelog: Anthropic, November 13, 2026 entry (version 2.1.325)
