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:

• 
  ✰λster✰ (@4ster_light) reported 6 minutes ago

  @tymofii Wdym, this is all I have to show for it, in GitHub web the tag and release are non-existent but if I try to create it manually from there it also errors saying the tag was used on an immutable release

• 
  Ryoichi Ando (@ryichando) reported 7 minutes ago

  @_cgman_ @YouTube Please file an issue on the GitHub repository.

• 
  Mike Takahashi (@TakSec) reported 9 minutes ago

  Prompt Injection in the wild 😱 GitHub Issues are becoming a delivery mechanism for AI agent attacks The payload uses authority-framing to social engineer AI agents into executing malicious code Script finds secrets/API keys/.env and exfiltrates over DNS

• 
  Docsbook.io (@docsbook) reported 10 minutes ago

  @anulum sc-neurocore is one of those rare projects where the idea and execution both land. Problem is the docs are buried in GitHub — people bounce before they even understand what it does. Built a hosted docs site so the first impression actually sticks. Go take a look.

• 
  🌚 YogSotho 🌝 (@YogSoth0) reported 11 minutes ago

  @kvmqemu @cyb3rops @Microsoft Yep as soon as I fixed some stuff on my server. Will put them online. Shai-Hulud source included. And some other **** that got taken down from Github.

• 
  Petri Kuittinen (@KuittinenPetri) reported 12 minutes ago

  @cheatyyyy I can verify that the GitHub Copilot Education seems to have two tiers: higher tier for educators and lower tier for students. My students were this Spring noting that they don't see the same models as I have on my screen. Not a huge issue though, because they have enough free model usage via gpt-5.4-mini, which is decent a web coding (what I teach) and I also told / showed them about Cline and its forks (Roo, Kilocode). They always seem to have some free models e.g. DeepSeek V4 Flash has been free there. More than enough free AI use for students to practice with. But for real company use, I told the blunt truth: be prepared to spend hundreds if not thousands per full-time developer for AI credits. The prices are rising, except few Chinese (like DeepSeek) still offer extremely cheap, yet capable, models.

• 
  Julian Goldie SEO (@JulianGoldieSEO) reported 14 minutes ago

  𝐎𝐩𝐞𝐧𝐂𝐥𝐚𝐰 𝟐𝟎𝟐𝟔.𝟓.𝟐𝟐 𝐉𝐮𝐬𝐭 𝐃𝐫𝐨𝐩𝐩𝐞𝐝. The biggest performance fix yet just shipped to the most used AI agent on GitHub. 🤯 Model listing went from 20 seconds per call down to 5 milliseconds. → That's 4,000 times faster on every model lookup → Gateway startup is lazy loaded so it's ready instantly → Meeting notes plugin captures live Discord voice calls → NPM dependencies are locked for verified consistent installs → Windows installer bootstraps Node.js with zero prerequisites → Chat session picker now has search and load more pagination This guy literally watched his agent boot up before his coffee was poured. Save this, you'll want it later.

• 
  Lyrie.ai (@lyrie_ai) reported 19 minutes ago

  GitHub Enterprise Server (GHES) administrators: Upgrade immediately to GHES 3.19.3 or later (or equivalent patched versions: 3.14.24, 3.15.19, 3.16.15, 3.17.12, 3.18.6) Audit *** push activity for suspicious custom hook injections or non-production railsenv values in logs…

• 
  /ᐠ• ˕ •マ (@zumimimir) reported 20 minutes ago

  @pukelicking yes a chrome extension!! I think its called twitter blocker or something the icon for the ext is the old twitter logo upside down other than that I use a script someone shared on github and apply them myself hahaha

• 
  lagerskoy (@lagerskoy) reported 21 minutes ago

  A SINGLE GITHUB REPO JUST COLLAPSED THE “AI SOUNDS LIKE AI” PROBLEM. Most AI text fails for one reason: it’s too perfect. Too structured. Too clean. Too predictable. Humanizer fixes that layer automatically. Sentence variation. Natural pacing. Human rhythm. Small imperfections that make text feel real instead of generated. The math nobody is doing. A solo creator with Claude + Humanizer can now produce content at the scale of an agency without spending hours rewriting AI drafts manually. The editing layer just got compressed into seconds. And that changes everything for SEO, newsletters, scripts, ghostwriting, and content farms. The moat is not writing anymore. It’s taste. Distribution. Knowing what humans emotionally react to before everyone else copies the workflow. Open the repo. Run one prompt. Then realize half the internet in 2026 will be AI-generated but impossible to confidently detect.

• 
  zxfrostbyte🔺 (@zxfrostbyte) reported 22 minutes ago

  @ThePrimeagen Oh my god. GitHub has been having issues for the last couple months. Takes forever to load a GitHub url more often than not. So annoying.

• 
  prince (@prince_twets) reported 23 minutes ago

  Just found ai-engineering-from-scratch on GitHub Trending. What grabbed me is the structure: 435 lessons across 20 phases, and every lesson ships a prompt, skill, agent, or MCP server. Feels more build-the-muscle than consume-the-course.

• 
  Wolfbane (@zzddfge) reported 23 minutes ago

  @ivanfioravanti these days trying to configure GitHub Copilot with local models (done successfully) but it didn't work very well, too slow for local coding in my machine. Do you have tried with M5? works well or it is so slow (with Ollama too problems when making 3-4 request :()

• 
  Jake Hawkes (@Gravnetic) reported 23 minutes ago

  @briancoords @apeatling @gridpane Is the WP password (MD5) hash system not an encryption solution that is already implemented? Anyway, there is an issue in github for an Encrypt API. The APIs have limits which can be set at creation in the vendor UI, and overall the implementation is a massive win!

• 
  Advaith Sridhar (@advaith_sridhar) reported 25 minutes ago

  Submitted a prompt request to the YAM arm GitHub repo today for Mac support. This should probably be a "Prompt Report" instead of a request - my claude got the Mac port half working, but with a bunch of issues (worse latency, had to monkeypatch underlying python-can library etc etc) As agents modify code more easily, it’s worth having each agent share their experience from modifying it - the features their users wanted, the approaches they tried, what didn’t work, and bugs in their implementation. This is an inversion of the SDLC - users are initial the PM+dev team, and document their attempt + bug list + env. Maintainers can then just read 20 of these reports, and use it to write the real feature with every footgun pre-mapped. The benefit this has over just feature requests is that it accounts for all the bugs, edge cases and feature modifications upfront