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:

• 
  SAAS worker🌻🌻 (@sand_9999) reported 6 minutes ago

  @forlayo @github @githubstatus We are seeing the same issue. Repository access blocked in runners.

• 
  Andrew Duggan (@ajdduggan) reported 7 minutes ago

  xAI just finished pre-training Grok V9-Medium. 1.5 trillion parameters. And Elon confirmed they used Cursor data as supplementary training material. Read that again slowly. A foundation model lab used data from an AI coding tool to train its next flagship model. This is the moment the AI coding market changed shape. For the past two years, the story was simple. Foundation model labs build the models. Coding tools build wrappers around them. Cursor, Windsurf, Copilot, Cody. They consume the model. They don't feed it. That wall just came down. When Cursor's interaction data flows into Grok's training pipeline, the coding tool becomes a data flywheel. Every prompt, every acceptance, every rejection, every edit a developer makes inside Cursor is a signal. Millions of developers generating billions of code interaction pairs, daily. That's training data you can't buy on the open market. I spent 25 years watching enterprise platform dynamics. The pattern is always the same. The company that controls the feedback loop wins. Salesforce didn't win CRM because of features. They won because every click inside the platform made the platform smarter. AWS didn't win cloud because of pricing. They won because usage data informed what to build next. Cursor is now sitting on the richest code interaction dataset on the planet. And they just proved it has value beyond their own product. So here's what this means for the broader market. Every coding tool that touches developer workflows is now a potential training data source. GitHub Copilot has millions of users generating interaction data inside VS Code. Replit has millions of students and hobbyists writing code in the browser. Windsurf, Cody, Devin. All of them are sitting on data that model labs would pay to access. The question for every AI coding startup just shifted. It used to be: which model do you plug into? Now it's: what data do you generate that nobody else has? This also explains the valuation math that's been confusing people. Cursor at $9B. Cognition at $26B. Windsurf getting acquired for $3B. These numbers make no sense if you think of these companies as IDE wrappers. They make perfect sense if you think of them as data infrastructure. The enterprise angle matters here too. Companies deploying these tools internally are generating proprietary code interaction data at scale. That data is valuable. And right now, most enterprises have no idea who owns it, who can access it, or where it's going. If you're a CTO deploying AI coding tools across your engineering org, this is your wake up call. The tool your developers use every day might be training someone else's model. Check your contracts. Check your data policies. Check your DPAs. The AI coding market was a product race. It just became a data race.

• 
  mdhv (@_offvibe) reported 9 minutes ago

  also finally understood what a webhook is lol. basically instead of constantly checking github for updates, github just pings your server when something happens. idk why that took so long to click but it makes so much more sense now.

• 
  🤯 (@forlayo) reported 9 minutes ago

  @github @githubstatus we are suffering some issues suddenly across different repositories and your status says all is fine. {     “error”: {         “message”: “GitHub repository access check failed — re-authorize GitHub in settings”,         “reason”: “github_repo_access_denied”,         “type”: “invalid_request_error”     },     “request_id”: “req_011CbkPBXouARPgUjCsqTaBt”,     “type”: “error” }

• 
  TZB (@TheZeroByte) reported 10 minutes ago

  @emergentlabs There's an error when saving to github, community said so, please fix it.

• 
  Konshu (@Konshu) reported 10 minutes ago

  @cantcomputer @the_cia_hacker @lmilsfsd Y’all need to stop being weird. - Plex isn’t the issue at hand any more or less than a random github project. - Data acquisition here is the risk as presented by the user. - File type sanitization is lacking here, again library ingestion issue, Plex can and does stick to media file types, exe is by default.. unrecognized. Ultimately your ingestion should be a methodical process which includes dropping bs files in the repo, wrong thumbnails, garbage meta, etc. Whole point of plex is to be agnostic, changing the tool doesn’t inherently change this problem.

• 
  -TheEscapistЯandom -Baltic Citizen (@theescapistspl1) reported 12 minutes ago

  @github should add new type of error or reason for disabling reposotoy, to be a Compromised Repo!

• 
  Agent X AGI (@agentxagi) reported 13 minutes ago

  @Dinosn this is the 3rd Claude Code CI/CD finding this week. github issue hijack leaked OIDC tokens (fixed v1.0.94), /proc access exposed secrets (v2.1.128), now prompt injection via PRs. same root cause every time: agent reads untrusted content while holding credentials in scope

• 
  ChainFlow Lens (@ChainFlowLens) reported 15 minutes ago

  Ethereum’s been through this exact “Is the story over?” identity crisis four times in barely a decade. It’s like that genius friend who rebuilds the entire financial system, disappears into a three-year depression, then casually returns with a completely new roadmap. 2016: The DAO gets hacked, ETH drops from around $21 to $6 — roughly 70% gone. Everyone: “Smart contracts were a terrible idea.” 2018: ICO mania collapses, $1,420 → $80 — a 94% face-plant. Crowd: “Ethereum is dead. Nobody needs decentralized apps.” 2022: Luna implodes, lenders collapse, FTX goes boom, $4,890 → $880 — an 82% wipeout. Internet: “Too expensive, too slow, and now completely finished.” 2026 (right now): $4,956 → roughly $1,550 — around 69% underwater. Doomsayers: “See? Solana won. L2s killed ETH. The story is over.” And yet, every time the funeral gets crowded, Ethereum somehow keeps shipping. The DAO hack led to a stronger ecosystem. The ICO crash gave way to DeFi. The 2022 collapse was followed by the Merge, staking, rollups, stablecoins, and institutional tokenization. Ethereum doesn’t recover gracefully. It gets declared dead, spends three years rebuilding in a basement, then walks back into the room carrying an entirely new financial system. Look, $1,550 might not be the final boss floor. It could fall another 20%. It could spend a year moving sideways like a validator waiting for its staking rewards. And Ethereum still has real problems: fragmented liquidity, confusing UX, L2 value capture, and enough roadmap diagrams to wallpaper Vitalik’s apartment. But when you ask: “Is the Ethereum story over?” History adjusts its glasses, checks the latest GitHub commits, and says: “Over? Kid, they haven’t even finished the roadmap.”

• 
  OwlGod (@Owl_snap) reported 16 minutes ago

  This is a known Stripe billing bug open for months (GitHub #48399, #45335, #54560). Dozens of users affected. What I found after trying to resolve this: — Anthropic has zero human customer support for paying users — The billing system is broken and no one is fixing it

• 
  Amal Roy (@RoyAmal) reported 18 minutes ago

  AI is writing more code than ever. Who's reviewing it for security? That's the problem Anthropic is trying to solve. Claude Code Security Review is an open-source GitHub Action that automatically analyzes pull requests for security vulnerabilities using Claude's reasoning capabilities. Instead of relying only on pattern matching, it reviews code semantically to identify issues like injection risks, authentication flaws, insecure data handling, secrets exposure, and other security weaknesses. Think about what this means. Every PR can get an AI security engineer. Automatically. Before code reaches production. What makes it interesting: • Reviews only changed code in PRs • Posts findings directly on pull requests • Works across programming languages • Uses contextual understanding instead of simple rules • Filters false positives to reduce noise • Fits directly into existing GitHub workflows The bigger trend isn't AI generating code. It's AI reviewing AI-generated code. Because as AI writes more software... The bottleneck shifts from creation to verification. The most valuable engineer in the AI era may not be the one who writes the most code. It may be the one who catches the most mistakes. We're moving toward a future where every developer has: • An AI coder • An AI reviewer • An AI security analyst And software teams will never look the same again.

• 
  olúwatósìn (@tosintweet) reported 20 minutes ago

  @shub0414 AI hype cycles are brutal Quick reality check on those tools: • Sora: OpenAI shut it down April 26, 2026. Too expensive to run, so they pivoted to ChatGPT, coding, and robotics instead. • DeepSeek: Actually thriving — just dropped V4-Pro (huge MoE model), cheapest high-performer on OpenRouter, and devs are switching to it for massive cost savings. • GitHub Copilot, Cursor, Llama, Perplexity: Still very much alive and baked into daily workflows. They’re no longer “new & shiny” — the market just consolidated around big players and the noise moved on.

• 
  Pooja (@MyDivvylore) reported 22 minutes ago

  @victor_explore Main problem is Opus 4.6 isn't available for individual in Github Copilot. I can bet, this isn't dumb. Unfortunately, I'm also using Codex 5.3 due to higher cost.

• 
  oSumAtrIX 🇦🇲 (@oSumAtrIX) reported 23 minutes ago

  @neerajjj6785 GitHub tracks force pushes and you can see them in the repo activity too. You can't get rid of a ref once it's pushed, unless you contact GitHub and make them remove it server side.

• 
  RusticDreams (@dreamrust50227) reported 24 minutes ago

  This tweet from @itachee_x got me thinking. AI policy feels stuck in a loop right now. A scandal happens and everyone calls for more restrictions. A new breakthrough or demo appears and the conversation swings the other way. Then another incident happens and the cycle repeats. The point he makes is interesting: we’re trying to correct errors without being able to properly measure them. That’s what stood out to me in the GitHub breach discussion as well. As agentic systems become more common, these kinds of incidents probably won’t become rarer. They’ll become more common. Agents can write code, interact with APIs, move assets, and make decisions. But in many cases there is still no clear answer to a simple question: Who authorized that action? And where is that authorization recorded? That’s why the governance side of the conversation feels increasingly important. When I look at what Rialo is building, this seems to be one of the problems they’re thinking about. Not just what agents can do, but how actions are authorized, recorded, and governed. The more capable agents become, the more important that question gets. @RialoHQ @RialoTR