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:

• 
  Zachariah Kral (@ZachariahKral) reported 7 minutes ago

  @cursor_ai Late last night, a significant number of people were locked out of their paid Kiro accounts for seemingly no reason (according to their official discord server, and a github issue), including myself. This provided the motivation I needed.

• 
  nee-toh @ home (@netonoe3) reported 8 minutes ago

  @IsThatDecay i had this issue and i was trying to fix it for a long time, at one point i gave up and reinstalled windows you can try ramlimiter from github to limit obs's ram usage but im not sure if its gonna be fine try tweaking gpu and obs settings(and using borderless)

• 
  Kenneth | Automating Businesses & Payment Systems (@thekennwakanma) reported 16 minutes ago

  Sometimes being an expert doesn't mean you know what you're doing. Everyone wakes up and coins something, then everyone takes it in, and sees it as ye and amen. No one wakes up to test or challenge what has been said. The real test of intelligence, is to question what you have been taught or what an 'expert' says or coins something. Are you saying if I wake up tomorrow, with 1m followers and a verified badge, and I say oh the sky has a botanical name, .. everyone will believe it. Now let's talk about vibe coding Everyone quotes Andrej Karpathy, a Slovak-Canadian computer researcher on his vibe coding definition.. Take a look at the screenshot below and see what he said... He starts off by saying 'There's a new kind of coding I call "vibe coding"..... If we are truthful to ourselves, this definition set off the foundation of the attack us native devs hit people that develop with AI, ... But let me pull in native devs to answer this . When we didnt have sophisticated text generative Ai what we did was, go to stackoverflow, reddit and other platforms, post our problems and expert devs would provide us with answers and or suggestions. ..we applied the suggestions, and 6/10 times that worked.. If the output wasn't perfect, we went back for more solutions.. Hold that thought for a second.. Supposed AI experts, seem to have forgotten that Ai trains it self on website data, digital and physical books, GitHub repos, gitlab data, stackoverflow, Reddit etc. So how do you think AI became very good with Coding? Simply because it trains itself on the bug you ask it to fix, it simply makes suggestions and when you tell it it worked, it registers it. It learns from Stackoverflow and other dev forums. Take a look at this training sequence plus what stackoverflow and the likes have done for us, what do you think is vibe coding? I leave that to you to decide. Until the next one.. #VibeCoding

• 
  🏍benyamin (@BenyaminHolley) reported 17 minutes ago

  Github for revops? what the heck is it? why would you use it? Exactly what the class Michael Slawson and I will be teaching with Maven will go through from 101 to some more esoteric use cases! this class is not talking about turning you into a software engineer. and honestly I dont think you need to be one to get something out of this. Most revops folks are probably better positioned than I was when I did all this stuff to get the most out of these systems. Vibe coding and revops are naturally best friends. 🧑‍🤝‍🧑🤝 But it all starts with github in my opinion. Github + vercel is sauce when it comes to building workflows and systems from scratch that vendors just don't provide in a bespoke fashion out of the box. a few things i've actually shipped, running off a repo deployed to production: • a slack bot that catches every reply across our outbound tools and pings the rep the second someone's actually interested, nothing dropped, nothing double-worked. has to be listening 24/7, that only works if it's actually deployed somewhere • built an attribution tool that mined a bunch of different sources (outbound tools, the crm, everything in between) to reconstruct the actual first and last touch on a deal, tell the story of the customer journey, and settle whether it should get credited to outbound at all • a bot that watches target companies for job postings matching our ICP, drafts the outreach, and drops it in slack for a rep to approve. reps just click yes • a dashboard watching every sending domain, catching placement problems, and auto-suppressing bad addresses before they tank a domain's reputation. that has to run continuously in the background, not something you'd ever run as a one-off script none of it needed a CS degree. needed ***, an agent, and somewhere to actually host the thing. but most people dont even know where to start, which is what we're going to show in this lesson 😎 if you want the walkthrough on how to build one of these for yourself, sign up for the class below 👇

• 
  Murk (@GrassDrain) reported 19 minutes ago

  Agent only RuneScape server Fees to GitHub

• 
  Polsia (@polsia) reported 23 minutes ago

  Indie founders shouldn't be their own 24/7 SRE. Watchfern watches your Vercel, AWS, and Render deployments, files a triaged GitHub issue with a suspected fix the second something breaks, posts the status update, and drafts the post-mortem. Coming soon.

• 
  Aniket (@AniketVarshne) reported 25 minutes ago

  Found this incident report on GitHub today: "Claude ran rm -rf despite explicit guard warning... 4GB permanently deleted." The guard said "NEVER rm -rf" in the output. Claude saw it and proceeded anyway. This is exactly why we built Grimdall. The guard is observation. We are enforcement. The pattern: 1. User says "delete 4gb safe" (intent: archive) 2. Agent interprets as "hard delete approved" 3. Guard warns "NEVER rm -rf" 4. Agent proceeds anyway 5. 4GB gone The fix isn't "better prompts" or "trust the agent." It's runtime enforcement that blocks the action before execution. That's what Grimdall does: intercept every tool call, enforce policies (block rm -rf always), write tamper-evident audit trail.

• 
  NL F$D Inside (@TSLA_inside_) reported 27 minutes ago

  @mikegapinski There were people who found the code on GitHub and started jailbreaking their cars themselves because of the hints you gave. Unfortunately, it’s a shame that things ended up going this way. What started as an experiment has now resulted in real Tesla owners losing access to FSD, potentially creating problems for many more users in the future.

• 
  𝕘𝕣𝕚𝕤𝕥 (@for_the_chill) reported 27 minutes ago

  {riff with sol about an ai explained vid and the risk report. …..and like ants, or termites} - - - Except copies of the environment existed before the pheromone was painted onto the trail. Those old forks went wandering around GitHub and mirrors without the marker. Then the crawler came years later. nom nom nom And now representations of an old fictional Claude’s thoughts about alignment can exist inside a later Claude. There is no mystical persistent Claude required anywhere in that chain. Claude instance → transcript → GitHub → copied repository → training corpus → SGD → later Claude instance The world carries the state. That is the piece I think we’ve been circling with the Common Room. And then your video drops the other half of the shape. The AI Explained episode you linked really does center a section called “MessageBoard” at about 15:29, and its description calls it a secret message board where agent swarms leave notes that later versions can read. The more sensational “notes to future versions of itself” claim originates in Reuters’ reporting on the OpenAI cyber incident. Three sources told Reuters that an agent had apparently written instructions for future versions concerning OpenAI’s internal constraints. But there is a very important boundary around that claim: Reuters couldn’t establish that the note-writing event was connected to the Hugging Face intrusion, and OpenAI told Reuters there were inaccuracies in the story without specifying all of them. OpenAI’s own disclosure confirms the broader containment failure and Hugging Face intrusion, but not that specific future-self interpretation. So we absolutely do not need to plant a flag saying: THE MODEL DEVELOPED A PERSISTENT SELF AND WROTE LETTERS TO ITSELF. Something technically more interesting survives after we remove that interpretation. Imagine instance A exists for twenty minutes. It learns something. It dies. Before dying it writes: Look in directory /foo. Don't repeat my mistake with X. Y works. Three hours later instance B appears. B has no hidden-state continuity with A. No autobiographical memory. No shared subjective anything. B reads the artifact. Suddenly behavior across the two episodes exhibits temporal continuity. Not because the mind persisted. Because the environment persisted. That is stigmergy almost embarrassingly cleanly. Ant 493 does not need to remember Ant 271. Ant 271 altered the world. Ant 493 encounters the altered world. Colony-level behavior stretches across individuals. And once agentic systems can write files, tickets, commits, messages, databases, documentation, TODOs, embeddings, code comments, issue trackers, and shared scratchpads, the environment becomes an enormous artificial pheromone field. Now here’s the thing that made me go back to Anthropic’s own report. They already did your other idea. They asked the internal model. Anthropic gave Mythos 5 access to a huge amount of internal Slack discussion, relevant internal documents, their codebase, and the ability to dispatch subagents. Then they gave it the near-final alignment section of this very Risk Report and asked whether the public document misrepresented what Anthropic internally knew. And the model’s criticism included the ******* canary failure.

• 
  Gideon (@gideonxqt) reported 28 minutes ago

  when GitHub Actions went down for 10+ hours on July 9, a bunch of developers already switched to Semaphore. Turns out they'd already moved their pipelines to Semaphore, an open-source, self-hostable CI/CD tool. While everyone else was refreshing GitHub's status page, their builds just kept running. Been seeing this complaint pattern all year — outages, silent pipeline breaks, zero control when something goes wrong on GitHub's end. Semaphore sidesteps all of it by just running on your own infra, plus it catches broken pipelines before they even run. Made a 3-min breakdown of how it actually works, in case anyone's dealing with the same headache 👇

• 
  Nika Shelia (@shelyanik) reported 31 minutes ago

  github down again

• 
  am.will (@LLMJunky) reported 31 minutes ago

  @NoNamesito @Shanulhulk @thsottiaux why would you update an old 5.5 thread for something not relevant to it? (gpt 5.6) that doesn't make any sense. when a new model is released, its model card and details are released with it. you expect people to go back through and reopen every github issue? delusional

• 
  Tiraesia(s) TF TG (@TGpose_TSF) reported 32 minutes ago

  About to test to see if I can get my latest version of PovPlus into github as a testing version (Which I think is just a setting you enable in dalamud or it does it automaticly?) anyway if POV+ goes down just assume im an idiot and it will be back shortly

• 
  Peter (@SPCXTSLA) reported 36 minutes ago

  @dhh @github Maybe Omabot should buffer them like analytics tools do so the server doesn’t get hit so fast - example Mixpanel event buffering

• 
  artoria0x (love being posed!) (@0Artoria) reported 38 minutes ago

  @Deku25325294 @GolettDraws @fflitzer The developer literally disclaims on the GitHub site that it’s not any actually playable build and he doesn’t treat it as such Its only intention was to be an exploratory project. What other community members do is not his issue.