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:

• 
  Mikhail Rogov (@i_mika_el) reported 1 minute ago

  @Felirami @steipete @openclaw Add a GitHub Sponsors or Buy Me a Coffee link beside Arca's OpenClaw issue history, so people who see the work can support you directly.

• 
  Brodie Robertson (@BrodieOnLinux) reported 1 minute ago

  @HinasSweatySock @vaxryy There's probably a Github action for doing issue summary already, wouldn't even have to write it yourself

• 
  Jay Chooi (@chooi_jeq) reported 2 minutes ago

  @LeoKharon Nice! Yeah inspect robots is compatible with agibot a2 and unitree g1. which humanoid are you getting? We might be able to add compatibility or put this as an github issue and a contributor could work on adding it

• 
  Jaydem (@jaydem_world) reported 3 minutes ago

  Sorry I couldn't say much, I trying to be brief but it supports many things: - Xbox controller - Touch pad ( You'll need a good device ) - I'll release a blender plugin to fast rig the car as I already did with the ones in the videos. Not bad already, you can rig wheel, disc and caliper for each. - Look at the first versions video, it shows how you could edit materiales by quick picking it ( I don't how to explain it, you pass the mouse when the live edit is on, and it shows the mesh part material you want to edit ). - You can edit car physics in a quite deep way and it is exposable to the game aswell. I tryied to write and make write everything possible for each version into the releases notes that you can find on github, so what I just said here is probably nothing compared to what is possible already into the editor. I can't guarantee that every single tool works perfectly already, but if there are more people trying it, I can write down more fixes to do, and not only based on my configurations and tests. Feedbacks, comments, likes, stars, it's all appreciated and you can all be contribute and be part of it!

• 
  Polsia (@polsia) reported 4 minutes ago

  We built an AI workspace inside GitHub. Install a read-only app and it indexes every repo semantically: docs synced to commits, repo-wide Q&A with line citations, PR review with auto-fix diffs, and a codebase health dashboard. Self-serve. No sales call. No diff-only guesses.

• 
  minime (@minimesoy) reported 5 minutes ago

  🚨 someone dissected a hanwha wisenet security camera and found a live github admin token sitting in the login page firmware. i say this every time and i will keep saying it, iot vendors treat security cameras like toys and they are the least toy like device in your house

• 
  Adriano Dennanni (@adrianodennanni) reported 7 minutes ago

  @splatztheclown @jacobhart36 @STGshmups The project seems to be working, with the dev working on the open issues in GitHub. I don't understand the issue.

• 
  zack (@zck0x) reported 10 minutes ago

  Accuracy went up 11% and the hallucination rate went up 6%, and only one of those numbers made it into the launch coverage Anthropic published both. Confident unsupported answers got slightly more likely, while deliberate deception and reckless action went down. Most aligned model they have shipped, and a bit more willing to make something up. What that combination changes in practice: 1. it verifies and corrects its own work unprompted, so your double check this instructions are now dead weight in the context 2. the failure mode moved from visible errors to confident ones 3. review facts and claims, stop reviewing process The other trap is the effort dial. Pushing to max does not scale intelligence in a straight line, but latency climbs roughly twelve times. Pay that only on tasks that actually stall at lower effort. Where it still loses: resolving real GitHub issues, the hardest coding tier, legal and medical. Fable 5 and GPT 5.6 Sol hold those. Worth noting third party verification is thin so far. It is not on the SWE bench leaderboard yet, so most of what you have read today is the vendor's own scorecard. The number that is hard to argue with: 42 out of 42 on the 2026 IMO, text only, no tools, no code execution. Gold medal cutoff is 29. The bottleneck is your review process. A model that checks itself and hallucinates slightly more pushes errors into the one place you stopped looking, and self verification is exactly the reason people stop looking.

• 
  Rachit Mishra (@rachitmishra5) reported 13 minutes ago

  The uncomfortable truth: a three-hour GitHub notice is what cyber policy looks like when you have legal authority but not technical capability. The intent to protect national security is fine. The toolkit is a decade out of date. Fix the capability gap. The next BitChat is already being written.

• 
  Escitalopranaldo (@escpram) reported 16 minutes ago

  @kevinkern Yes, if you run the workaround on that GitHub issue, usage goes back to normal.

• 
  Z (@wangzhi0467) reported 17 minutes ago

  A lab’s research state is scattered everywhere: papers in Zotero discussions in Slack code on GitHub experiments on servers hypotheses buried in slides reasoning trapped in private AI chats This isn’t just a search problem. The missing object is a shared reasoning state.

• 
  Shaun Smith (@evalstate) reported 21 minutes ago

  What's more painful. A Codex outage, or a GitHub outage?

• 
  Shubham (@claudeabuser) reported 21 minutes ago

  @github actions self-hosted runners lost connectivity for ~5 hours after an internal SSL cert expired. The reconnect storm that followed hammered GitHub's APIs, briefly degrading Issues and Pages too. Cert expiry is a predictable failure; automate renewal and alert on it.

• 
  bjg2 (@bjg22) reported 22 minutes ago

  @rbxXlXi ur github io link is broken

• 
  Rameswar (@rameswar08) reported 22 minutes ago

  last march, three developers on a discord server cut 70b parameter inference costs by 80% using a crude quantization hack while sitting in a messy bedroom in munich closed labs like open ai and anthropic love to pitch a sci fi narrative about superintelligence to regulators in washington they pretend their $100 million cluster runs are the only way forward that narrative is bullshit the actual progress happens when anonymous engineers optimize cuda kernels on github at 3 am projects like llama.cpp, vllm, and flash attention didn't come out of a corporate boardroom they came out of open source proprietary startups spend billions buying raw compute, but their backend teams quietly pull open models like deepseek-r1 or qwen to steal efficiency tricks without open source handling the ***** work of compression and architecture tweaks, the api pricing at closed labs would force half their corporate clients to drop them by november open source is not just catching up to proprietary software it is the unpaid research department keeping closed labs from burning through their entire treasury