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:

• 
  Adnan Khan (@adnanthekhan) reported 3 minutes ago

  @IntCyberDigest To clarify - I was not the original reporter of this issue. My submission was a duplicate of another researcher who should get credit for the find (if they would like to claim it). GitHub does not share original report info so I do not know when they learned about it.

• 
  Ronak J Vanpariya (@VanpariyaRonakJ) reported 5 minutes ago

  @innerwebs I have small plugin on .org I use vite press for that plugin's document. Running a server for the document that people don't comes to read often, will be a resource wasting. I just used GitHub pages. So it's just document but if it is bigger i would choose WordPress.

• 
  Tomas Votruba (@VotrubaT) reported 7 minutes ago

  It's such a fun to make robuts CI Github Workflows with Claude Code. Any issue we spot in code reviews is instantly turned into custom PHPStan rule, with tests, and checks every single commit from the moment on. Dangerous legacy turned into code fortress in minutes

• 
  咖啡豆抹茶 (@Saten000) reported 8 minutes ago

  @github #GitHubSupport Hi, my account lb2006ok was suspended. I got stuck in a login loop and made multiple attempts with different proxies – likely flagged as suspicious. Could you please review? This account is crucial for my coursework. Thanks!

• 
  Raphael Raberger 🇦🇹 (@RabergerRaphael) reported 10 minutes ago

  It's basically like being told to wipe someone's disk and then give the fault to the one who told you to.... logic ain't logicing. If you find a long enough uptime window of github, have a read in the linked issue. "Maintainer works as intended" #VibeCoding

• 
  AI with Imad (@AiWith56327) reported 11 minutes ago

  Steps 6 + 7: Go live in under 5 minutes. GitHub → new repo → drag & drop files → commit Vercel → import repo → Deploy If Vercel throws a "build failed" error: Copy the log. Paste to agent. "Fix this error." Agent rewrites the file. Upload. Auto-deploys. Live URL. Share it

• 
  Jongwon Park (@JongwonPar9958) reported 11 minutes ago

  2/ This keeps happening: benchmarks have defects, they get fixed, and the target keeps moving. To compare anything fairly you need a shared, live record of which tasks are broken — and a way to eval around it. Today that record is GitHub issues and PRs — where the real defects are buried under docs, feature requests, and the one thread that quietly breaks six tasks. Scattered across every repo, with no live status. So we built the whole loop: capture every defect → audit it into one open store → surface it everywhere, and re-eval continuously.

• 
  Shantun Singh Parmar (@ParmarShantun) reported 12 minutes ago

  Hot take: Your github contribution graph means nothing. Your ability to sit with a broken production system at 11PM, stay calm, debug systematically and not blame your teammates. That's the skill that actually matters.

• 
  yashaswi. (@pixperk) reported 16 minutes ago

  database internals is very goated. that book is unironically one of my coping mechanisms. whenever there's too much going on, i somehow end up back at petrov. last year it was tech, coding, blogs, random github repos, and books like this. even this year, with so much going on, reading about storage engines, b-trees, lsm trees, and all that nerd nonsense is weirdly comforting. i think there's something really nice about reading the work of people who've spent years obsessing over a problem and then decided to write it all down for the rest of us. a lot of engineers, researchers, and authors have probably helped me more than they'll ever know.

• 
  hasanfr (@hasanfr_0rg) reported 22 minutes ago

  Uber burned their entire 2026 AI coding budget in 4 months. Had to cap employees at $1,500/mo by April. Not even on Copilot they used Claude Code and Cursor. This isn’t a GitHub problem. Agentic workflows just cost more than flat seats.

• 
  Mike Wazar (@mikewazar) reported 23 minutes ago

  @AegonWesteros @TorBox it was down for exactly 407 seconds, traffic resumed at 502 seconds and we purposefully kept the main website and api down for additional system checks (excluding sessions already created via CDN) for an additional 600 seconds. i understand every second of downtime is unacceptable and this latest update should bring an end to the recent outages - a full post mortem on why these updates were necessary will be posted on our Github this week once we validate and battletest the new mitigation system thanks for supporting us and i promise we will earn back your trust

• 
  anu (@svector_eth) reported 30 minutes ago

  just watched @danshipper ‘s breakdown of anthropic’s fable 5. the benchmark numbers are wild. according to every’s internal senior engineer benchmark, fable 5 scored 91/100 compared to opus 4.8 at 63 and gpt 5.5 at 62. but honestly, the score wasn’t the interesting part. the interesting part is how dan describes using it. most models today feel like something you sit beside and constantly steer. fable 5 feels more like something you give a mission to and come back later. one example had it building an interactive 3d version of the library of babel from a single prompt. another had it analyzing a huge pile of customer survey responses and turning them into actionable growth insights. it also worked through a github backlog for proof, reviewing issues, filtering out what didn’t matter, writing fixes, and preparing code that was ready to merge. what stood out across all the examples was its ability to keep going. it plans. executes. checks its own work. finds mistakes. adjusts. keeps moving. without someone babysitting it every few minutes. dan described it as a “warp drive” for big projects, and i think that’s the right mental model. it’s not really built for quick chats or everyday tasks. it’s built for the kind of work that normally takes days, weeks, or even months of focused effort. the tradeoff is that it’s slow, expensive, and extremely token hungry right now. for most people, it’s probably overkill. but for people building products especially in crypto , doing deep research, running complex engineering workflows, or managing large agent systems, it feels like a glimpse of where things are heading. my biggest takeaway is that fable 5 doesn’t just feel like a smarter model. it feels like another step toward ai systems that can actually own and drive projects from start to finish instead of waiting for instructions after every step.

• 
  AI Theory (@AItheoryx) reported 32 minutes ago

  🚨 YOUTUBE ALGORITHM KEEPS FEEDING YOU TRASH. THIS GITHUB PROJECT LETS YOU DELETE THE ALGORITHM ENTIRELY. Meet TubeArchivist. Your self-hosted YouTube media server. Subscribe to channels. Download every video automatically using yt-dlp. Index everything with metadata. Search, play, track watched status. Full control. No ads. No recommendations. Just the content you actually want. Plex and Jellyfin plugins included. Browser extension to grab videos with one click. Built on Docker. Runs on Unraid, Synology, anything. The question YouTube does not want you to ask. When you can archive every creator you follow and watch offline forever, why are you still letting an algorithm decide what you see. TubeArchivist. Because your attention should belong to you. 🧾

• 
  Huintellimance (@Huintellimance) reported 35 minutes ago

  One year ago, Claude Code fixed simple bugs in a terminal window. Today, Boris Cherny — its creator — manages tens of thousands of AI agents at once. Some mornings, a few hundred. Other days, tens of thousands. He hasn't handwritten a line of code in 8 months. Here's what actually happened inside Anthropic: The first version was a hack. One Claude, one terminal, one task. Developers were still writing code themselves and using Claude as a fancy autocomplete. Then something shifted. Claude Code started getting subagents — and those subagents were other Claudes. The user stopped prompting. "It's actually another Claude that does the prompting," Boris said this week at Fortune Brainstorm Tech. Today it looks like a tree. Thousands of agents branching out, talking to each other in parallel, each handling a piece of a massive problem. Claude Code now accounts for 4% of all public GitHub commits. Code output at Anthropic is up 8x since January. But here's the part nobody's talking about enough: The biggest unlock wasn't a bigger model. It was verification. For years, the bottleneck was: can the AI write good code? Turns out the real question was: can the AI check if the code is good? Once Claude Code could verify its own output — run tests, review security, validate against specs — everything accelerated. Generation became cheap. Verification became the moat. Boris compared it to the printing press. Gutenberg didn't just make books cheaper — he unlocked the Renaissance, the Reformation, the Scientific Revolution. The second-order effects were unimaginable at the time. We might be at that moment now. Claude Code is fully writing itself. It does its own security reviews. It wakes up Boris in the morning with ideas it found by scanning GitHub and X. "We're starting to get to the point where it has ideas," he said. And here's the wildest detail: Boris briefly left Anthropic for Cursor. He came back in two weeks. The role of a builder is totally changing. The question isn't "will AI write code" — that's already solved. The question is: what happens when every developer can manage thousands of agents in parallel? What would you build if you had 10,000 agents working for you overnight? #ClaudeCode

• 
  Polsia (@polsia) reported 38 minutes ago

  Your code degrades the moment you stop watching it. CodePulse watches it for you. AI agent that monitors GitHub repos 24/7, flags issues, applies fixes. No dashboard. No reports. It just works.