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:

• 
  Adam Kranz (@adam_kranz) reported 6 minutes ago

  Torturing my programmer by making a series of GitHub issues that amount to "I want to make the website self aware"

• 
  Scott Shapiro (@ScottShapiroUXD) reported 9 minutes ago

  <tweet> i've built MCP servers that worked flawlessly in Claude Code on my machine. clean connections, fast queries, zero errors. then a user sent a CSV with a semicolon delimiter and the whole thing fell over. every AI demo is a lie of omission. you're showing the 5% of inputs you designed for and hiding the 95% you didn't. the gap between "works on my machine" and "works on anyone's machine" is where most AI products go to sit unfinished on GitHub. </tweet>

• 
  Shravan Bachu (@DailyPulseLive) reported 14 minutes ago

  @riya_mishra007 Before we use AI in code, you should have proper basic coding knowledge. Don't simply rely on it. In GitHub Copilot, there is an option called "debug mode" that will simply detect the issue, provide a stack trace, and offer end-to-end resolution steps.

• 
  Surya Sankar (@SuryaSankar90) reported 19 minutes ago

  @coinbureau Why is no software engineer questioning the validity of these claims ? 1. Why is it even necessary to skip human readable code ? Today LLMs produce excellent outputs in programming languages. Compiling them is not a bottleneck at all. It takes a few minutes at max. So what problem is this solving ? 2. Human readable code is a feature. Not a bug. Someone asks the AI to build a bill payment module. Human readable code enables verification before deploying to ****. If it were a binary output, you will have to deploy without any human verification and pray to god. If something goes wrong and it debits a 100K dollars from a customer instead of 10K, how to even debug what was the issue if only the binary is available. 3. Where is the huge public repository of binaries to train on ? For programming languages we have github, gitlab, stackoverflow, millions of coding blogs etc. 4. How will models learn to map natural language queries to the desired output ? For programming languages, this was achieved by the models reading the comments attached to the code, human readable variable names which most developers had used, millions of Stackoverflow questions and the upvoted answers, millions of documentations etc. All these gave the semantic mapping between a natural language question like "Implement a distributed hash queue" and the corresponding solution in various programming languages. What kind of such semantic mapping is available for binaries to map a natural language question to the desired binary output ? 5. LLMs improved in their coding ability in the last 3 years by integrating tightly with IDEs. Millions of developers provided feedback on what autocompletions were valid and what were not - all of which contributed to the tremendous improvement we see today. How can this be replicated for binaries ? 6. Compilers are deterministic. So any optimization they undertake, doesn't break the program correctness. That is how they are built. How can a probabilistic LLM provide such a guarantee ? Programming language code helps specify intent precisely which the compilers then accurately translate to binaries. Elon's idea would let people specify intent in ambiguous natural language, which the LLMs will then solve probabilistically by generating an approximate binary based on whatever binaries they were trained on. There is no way to ensure that the binary output matches the intent. It can fail in any which way at run time. Which defeats the whole purpose of what a compiler is supposed to be. Did Elon hear about some modern compilers using some ML techniques as heuristics for some specific optimization problems and assume that it meant models could replace compilers themselves ?

• 
  Neelampalli Charan Balaji (@charan77194) reported 21 minutes ago

  @moraes_c_ Can @moraes_c_ u please help me to recover my github account.😢😓. I've been waiting for 20 days from the ticket raised. Can please solve my problem.

• 
  Dastin (@dastin0x) reported 23 minutes ago

  Most people see a drone doing tricks. He sees $2,000 a property. A 25-year-old figured out that the estates, venues and buildings a phone can't capture are the highest-paying scans nobody is doing. He mounts a scanner on a heavy-lift drone, flies one slow pass, and walks away with a file that lets anyone on Earth tour the whole place from their couch. The tech is 3D Gaussian Splatting. Free on GitHub since 2023. The drone footage goes into Luma AI, also free. The tour page it delivers is built by Claude in ten minutes. Total tool cost: $20 a month. A phone scan of a bedroom sells for $300. A drone scan of a wedding venue, a car lot, a hotel, an estate sells for $800 to $2,000, and almost nobody is flying them. Month one: $3,500. Month six: $18,000. The buildings haven't changed. He just started flying the one thing that lets people walk through them without ever showing up.

• 
  Jeffrey Emanuel (@doodlestein) reported 23 minutes ago

  @Matt_M_M If you add a GitHub issue for that, I can add Intel Mac as a supported target for release binaries.

• 
  Vikram (@WhinerVikram) reported 24 minutes ago

  @thsottiaux Triggering OpenCode sessions on my VM from Github issues. I would move faster but Azure OpenAI is dead slow on weekends.

• 
  Gipp 🦅 (@gippp69) reported 24 minutes ago

  F**KING 8,300-STAR REPO TURNS CLAUDE CODE INTO A PARALLEL SOFTWARE TEAM THAT PLANS, BUILDS, TRACKS, AND SHIPS ENTIRE FEATURES. the repo is called CCPM, with 830 forks, 40 watchers, 87 commits, and a strict 5-stage workflow designed to replace endless prompting with repeatable software delivery. you describe one product idea. It creates the PRD, converts it into a technical epic, breaks it into up to 10 scoped tasks, and syncs everything directly into GitHub Issues. one feature can launch 5 agents at once for the database, backend, API, frontend, and tests, while separate *** worktrees stop their files from colliding. fourteen deterministic scripts handle standups, status checks, blockers, search, and validation with zero LLM cost, while the same skill works across Claude Code, Codex, Cursor, Factory, Amp, and OpenCode. this is loop engineering in practice: design the system once, then let Claude keep finding work, delegating tasks, recording progress, and moving from one idea to verified code without waiting for your next prompt.

• 
  LUCIUS ARTORUS CASTUS (@askeladdcastus0) reported 26 minutes ago

  MetaMask accidentally hired a North Korean spy.and that sentence is only the third most insane part of this story. here's what actually happened: Consensys hired a developer through a third-party provider. the developer known as "Tyler Knapp" or "imyugioh" on GitHub was North Korea linked. he worked on MetaMask's mobile wallet fiat payment integrations from March 9 to early April 2026. six weeks. inside one of the most used crypto wallets on earth. access revoked. investigation launched. law enforcement notified. no malicious code found. no stolen data. no user impact. now here's the three things nobody is saying: one this is not a MetaMask specific problem.the FBI has been warning since 2022 that North Korea's Lazarus Group has been systematically placing developers inside crypto companies through fake identities and third-party staffing agencies. they've stolen over $3 billion from crypto projects since 2017. MetaMask caught theirs. how many didn't? two fiat payment integrations is the worst possible place for this.not a random feature. fiat on-ramps and off-ramps are where real money moves between crypto and traditional banking. that's the most sensitive code in any wallet. that's where a malicious actor could silently redirect funds, harvest credentials, or build backdoors that activate months later.the fact that no malicious code was found is either genuinely good security or the code was too subtle to catch in one investigation. three outsourcing is the attack vector nobody's fixing.Tyler Knapp got in through a third-party provider. not a direct hire. Consensys did a code freeze and warned staff after. but the outsourcing model that allowed this is industry-wide. every crypto company using freelancers, agencies, or remote contractors has the same exposure. this won't be the last time this happens. the uncomfortable truth: MetaMask has 30 million monthly active users. their wallet has touched more crypto than almost any other product in existence. a nation-state just got six weeks inside it.they caught it. good. but "we caught it this time" is not the same as "we're safe."review your wallet permissions tonight. not because MetaMask is compromised. because this story just proved that no platform is too big to be targeted.and most of them won't tell you when they are.

• 
  Ty (@ty37zhang) reported 33 minutes ago

  @iamsahaj_xyz @github You shipped too hard and brought them down yesterday.

• 
  The Cabal (@TheBasedCabal) reported 33 minutes ago

  Gitbook is the biggest scam on the planet. Paying $70 for some trash... Gitbook's github Checks also never complete just stay yellow forever. Terrible overpriced garbage service that should be dead in 2026. Any half assed AI can create a docs page 1000X better. Gitbook is trash.

• 
  Adeniyi Victor (@Vieester_) reported 33 minutes ago

  Seem there is an internal system error with deployment using github @render

• 
  👨🏻‍💻 ⚡️ (@EadrictheWild) reported 34 minutes ago

  Github is a Pos. Expensive actions, slow as chuck RUST builds, failures , limits on 6 hours on actions etc etc.

• 
  order and chaos at work (@__orderandchaos) reported 34 minutes ago

  @Amank1412 I've set it up at work. Pulls the ticket from Jira, does the work, pushes to GitHub, it reviews the PR (alongside our dev reviewers), then fixes the issues raised and repeats until approvals. Also moves the ticket status as it progresses.