GitHub Outage Map

Community Discussion

Tips? Frustrations? Share them here. Useful comments include a description of the problem, city and postal code.

Beware of "support numbers" or "recovery" accounts that might be posted below. Make sure to report and downvote those comments. Avoid posting your personal information.

GitHub Issues Reports

Latest outage, problems and issue reports in social media:

• 
  Broooooklyn (@Brooooook_lyn) reported 7 minutes ago

  @graykevinb @AMD @AMDRyzen Lol, they don’t even have an official setup-rocm GitHub action. They seem to expect developers to solve all the problems themselves, and then have everyone develop apps for them for free.

• 
  루이 (@_chiiazu66) reported 8 minutes ago

  @Fluffyquack If anyone is stuck on finding the update like I was, just go through the RE Framework github, the latest update is on there and it works perfect once you replace it with that one. Some mods may still be broken (the fov one I used needs an update for example)

• 
  Maverick | AI (@RizwanAly07) reported 9 minutes ago

  @Test_Sprite Open-sourcing the TestSprite CLI is a massive win for the AI agent ecosystem! 🚀 Giving coding agents the ability to independently test and fix their own code end-to-end solves one of the biggest production bottlenecks. Can't wait to try this out on GitHub!

• 
  Tacos and Airplanes (@blob_watcher) reported 10 minutes ago

  @teortaxesTex Copilot branding is terrible because it's being used as an umbrella term for a bunch of unrelated applications that happen to be hosted on MSFT servers. We have access to Claude and ChatGPT via GitHub Copilot via VSCode. Which is different from the in-GitHub Copilot. Terrible.

• 
  DFIR Lab (@DFIR_Lab) reported 10 minutes ago

  🦅 Tool Tuesday: Hayabusa — Fast Windows Event Log Analysis for Threat Hunters When you're knee-deep in a Windows compromise and staring at gigabytes of EVTX files, speed matters. Hayabusa is a Rust-based event log analyzer that rips through Windows event logs at scale, applying Sigma-compatible detection rules to surface threats fast. Built by Yamato Security, it ships with 4000+ built-in detection rules covering everything from credential dumping to lateral movement. It scans EVTX files offline, generates a consolidated timeline of security-relevant events, and outputs to CSV, JSON, or HTML — whatever fits your workflow. Real-world use case: You've pulled EVTX logs from 50 endpoints during an active IR engagement. Instead of manually parsing Security.evtx looking for 4624/4625 patterns, you point Hayabusa at the entire dataset. Within minutes, you have a sorted timeline flagging Mimikatz execution, suspicious PowerShell, and abnormal logon patterns — all color-coded by severity. Why it matters: Traditional EVTX analysis is slow. Hayabusa's Rust core makes it blazing fast, and Sigma rule compatibility means your existing detection content works out of the box. It's offline-capable, so you can analyze logs on an isolated IR laptop without network dependencies. Alternatives: DeepBlueCLI (PowerShell-based, lighter but slower), EvtxECmd (Eric Zimmerman's tool, great for parsing but less detection-focused), and Chainsaw (another Rust option with Sigma support). Get it: hXXps://github[.]com/Yamato-Security/hayabusa #DFIRTools #IncidentResponse

• 
  AClitheroeKid (@whitypedia) reported 11 minutes ago

  a sh!t end of the week, get product Devs to check their npm using a good resource, SOC flips out, found malware in sec tool, dev roll back their vms and we lock down devices, turns out after SOC reaching out to GitHub it was test files they forgot to defanged 🥺 ahhhhh

• 
  Snassy.icp (@SnassyIcp) reported 11 minutes ago

  Interesting conversation with @grok about the currently pretty depressing path we are on. I think we need better ideas than the current ways to handle increasing model capacity wrt finding security issues. Here’s one I have been pondering: Imagine if github added a ”self-healing repository” feature, where every time a big AI lab released a new model it would first run over all the self-healing repositories and fix the zero-days it discovers. Then, assuming all important projects opt in to self-healing, it would be safe to release the new model to the public without fear it would wreck havoc in the wrong hands.

• 
  JRC (@Jubleerc) reported 18 minutes ago

  From Enthusiasm to Caution- Enterprise AI story in 2026 Microsoft gave thousands of its engineers Claude Code in December. By June, it's cancelling most of those licenses. Not because the tool failed. Because the bill arrived. Token billing ate Microsoft's annual AI budget. Teams are moving to GitHub Copilot CLI by June 30. (The Verge) Uber's version was faster — its entire 2026 AI coding budget gone in 4 months. Power users: $500–$2,000 per engineer, per month. (Forbes) The mid-year scorecard: → 88% of companies use AI somewhere (McKinsey) → 2 in 3 haven't scaled past pilots → 95% of pilots show zero profit (MIT) → 40%+ of agentic projects will be killed by 2027 (Gartner) Everyone's using AI. Only Few are making money with it. But the other column of the ledger looks very different: JPMorgan: ~$2B/yr in AI value, matching its ~$2B spend. Dimon calls it "the tip of the iceberg." IBM: $4.5B saved using its own AI across 70+ internal workflows. Agents that survive pilot: ~171% avg ROI. Same models. Same vendors. Different discipline. That's the whole story. What the winners do differently: 1. Track cost per outcome, not total spend 2. Tie every project to real revenue or savings 3. Small models for routine work, big ones for hard problems 4. Humans in the loop on customer/money decisions 5. Give every pilot a kill date H1 didn't prove AI is overhyped. It proved AI is industrial — and industrial tools reward operators, not enthusiasts. The window to be early on disciplined AI is still open. The window to be casual about it just closed. What's your biggest AI lesson from H1 2026 ? #AI

• 
  Liam Castaigne 🔜 AC (@voird33r) reported 18 minutes ago

  @Code_Fault @LundukeJournal Yeah, they don't. They absolutely should. Package hosts are not taking this issue seriously. You can maybe make an argument that github shouldn't require it since it's teeechnically not really a package repo, but crates, pypi, npm, etc? I don't see the excuse.

• 
  Yep my name is Guy 😊🌸🥕 (@MyNamesGuy) reported 18 minutes ago

  @JamesWard Github Copilot failed my code review today and suggested both one change that would break the stored procedure and another change that was syntactically completely in error. It was so awful that I was wondering whether the LLM had been poisoned.

• 
  Khánh Phạm (@khanhphd) reported 20 minutes ago

  Reproduce: - download mac apps - open the app is alreay slow - click sign-in in the app - open the web which has buttons "Signin with Github" -> i click this (open this web is already slow) - and omg, just loading forever or click buttons don't respond

• 
  Traceback (@Tracebackqa) reported 26 minutes ago

  Green PRs still hide broken flows. - Traceback is the QA layer for teams shipping software. - AI drives the browser like a person; self-healing tests check every PR automatically. - Failures become work in GitHub, Linear, and Slack. Verify every product change before it ships.

• 
  泽林 🌫 (@zelin1107) reported 28 minutes ago

  1/ GitHub Actions not pinned to SHA Using @v6 instead of commit hash = supply chain attack risk. Fix: uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11 #GitHub #Security

• 
  Jolly Sampson (@Jolly69289037) reported 31 minutes ago

  @ireteeh real labs. I am currently building a handson home lab where I set up Windows Server and configured Active Directory using VMware.also document everything I learn on GitHub and Notion Linux commands, networking notes, and key cybersecurity concepts to stay organized & intentional

• 
  Ryan Delaney (@_rrdelaney) reported 34 minutes ago

  @jhasanofficial @karrisaarinen @linear Coding sessions run in a secure sandbox with no access to secrets or credentials, and limited GitHub access. Additionally, for externally created issues we lock down the sandbox's network access.