GitHub 在 2026 年发布的内容

GitHub Investigating Incident Affecting Git Operations Issues Actions and Pull Requests

GitHub experienced a service incident on October 7, 2026, impacting Git Operations, Issues, Actions, and Pull Requests. The issue was resolved within 47 minutes, with widespread impact from 16:52 UTC …

Purpose-built model for leaked secret detection

GitHub introduced a purpose-built AI model for leaked secret detection, integrating context-aware checks into secret scanning alerts, push protection, and GitHub Copilot security reviews. Existing AI-…

Local sandboxing for GitHub Copilot now generally available

GitHub Copilot now offers local sandboxing, a generally available feature enabling secure execution boundaries for agentic workflows on developers' machines. Tools and commands initiated by Copilot ru…

Discover local models in GitHub Copilot CLI

GitHub Copilot CLI v1.0.94-0 introduces /modelto discover and select local models from a running Ollama instance, alongside cloud models. Users must manually add and confirm models, which support tool…

GitHub Adds Claude Haiku 5.5 to Copilot for Quick Coding Tasks

GitHub made Anthropic’s Claude Haiku 5.5 generally available in GitHub Copilot, targeting fast, high-volume coding edits, terminal tasks, and subagent workflows. The model is accessible across Copilot…

Thomas Dohmke launches entire, decentralized GitHub alternative for AI coding

Thomas Dohmke, former GitHub CEO, launched Entire, a decentralized platform addressing AI coding tool limitations by hosting code across regional hubs for speed and reduced bottlenecks. It integrates …

The difference between Codex and GitHub Copilot: Which one should your company sign up for? The deciding factor was 'where you ask for work to be done'｜Codex Studio

A third-party analysis clarifies that GitHub Copilot operates within GitHub (via GitHub AI Credits) while OpenAI Codex is accessed via ChatGPT plans, with Codex also usable within GitHub via Copilot s…

GitHub launched a purpose-built AI model for secret detection that analyzes surrounding code to identify leaked credentials, including non-standard formats. The model powers AI-detected secret alerts

Claude Haiku 5.5 in GitHub Copilot

GitHub Copilot now includes Anthropic’s Claude Haiku 5.5, a lightweight model optimized for fast, high-volume tasks like subagents and terminal edits. Early tests show it matches Sonnet 5 performance …

GitHub Copilot CLI version 1.0.94-0 introduces a /modelto command to discover and select local models from a running Ollama instance, alongside cloud models. Users must manually add and confirm models

GitHub announced general availability of local sandboxing for GitHub Copilot across CLI, the Copilot app, and VS Code sessions. The feature restricts Copilot-initiated tools and commands to a secure e

Secret protection must scale with software

GitHub reports that one in three pull requests now involves an AI agent, accelerating code creation and secret exposure risks. To address this, GitHub introduces a fine-tuned ModernBERT classifier tha…

Update your IDE to restore agent activity in Copilot usage metrics

GitHub identified an issue where Copilot agent activity was misattributed in usage metrics due to IDEs moving agent sessions to the Copilot SDK without proper IDE identification. A fix is rolling out …

Stacked pull requests generally available

GitHub announced the general availability of stacked pull requests, allowing developers to break large changes into smaller, independent PRs for easier review and merging. Repositories using stacks sa…

Over Half a Million Credentials Leaked on GitHub Remain Active

Truffle Security found 543,699 valid credentials in GitHub repositories, with some active for over 16 years. GitHub’s push protection reduced supported leaks by 53%, but 199,843 credentials were commi…

Former GitHub CEO Thomas Dohmke brings entire to tackle outages

Thomas Dohmke, former GitHub CEO, unveiled Entire, a decentralized Git platform designed to mitigate GitHub outages by distributing code hosting globally. The preview launches July 8, 2026, with hubs …

Code scanning AI Scan enablement status in security overview

GitHub introduced a new feature in its security overview that shows AI Scan for pull requests enablement status. Organization and enterprise admins can now view enabled and not enabled repository coun…

GitHub’s ReviewBench puts AI code reviewers to the test

GitHub introduced ReviewBench, a benchmark for evaluating AI code review agents, available in research preview. It tests 219 pull requests across 19 languages, measuring precision, recall, and F1 scor…

GitHub Copilot weekly releases — September 28

GitHub introduced new Azure canvases in the Copilot app for resource queries, cost health checks, and Azure Functions skills, along with VS Code Copilot 1.140 updates. These features enable automated …

Building Git infrastructure for agent-scale development

GitHub is overhauling its Git infrastructure to support agentic software development, driven by a 2x surge in total Git activity (218.2B to 473.3B events/month) and 7.38B commits in September 2026 alo…