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:

• 
  Tom Ciszek (@Ciszek) reported 34 seconds ago

  @kdaigle A2A and Agent Client Protocol, GitHub Actions are broken. Call me.

• 
  Redchou (@R3DCHOU) reported 4 minutes ago

  @pantharshit007 @RhysSullivan Got banned in the beginning of March, they certainly corrected this problem… And the GitHub post is saying people got banned from Antigravity, Gemini, etc… not from all Google services…

• 
  Ilias Haddad (@IliasHaddad3) reported 5 minutes ago

  I made it open source, and many people love it and find it useful. I started iterating by listening to bug reports, user feedback, and requests on GitHub. After a couple of calls with people about Edit Mind. Users wanna desktop app to reduce the friction of setting up a Docker service and managing the server themself.

• 
  Zach Kamran (@Zach_Kamran) reported 6 minutes ago

  Another serious Github outage today. Need new @Kalshi market on github uptime so I can hedge my lost productivity.

• 
  Mihaithebest (@realmihai_matei) reported 7 minutes ago

  GitHub putting secret + dependency scanning into the MCP Server is a very good sign. If agents are going to touch real repos, security can't live after the PR. It has to be inside the agent loop: inspect, change, scan, fix, repeat.

• 
  hieu (@0xk2_) reported 11 minutes ago

  I am happily using codex with following plugins: figma, google mail, calendar, drive, github, superpower, hyperframe, remotion. I am not someone with excessive plugin installation; just average. However, when I dig deeper. Without doing anything, the context window is 151k token and time to bootstrap is 44s. Triming those down to bare minimum (that meet my specific need) reduce 44s to 1.05s. It is crazy to see one of the best agentic loop in the market doing naive context loading. To achieve what? the AGI feeling with the cost of efficiency and accuracy. I have very very very high hope on @openclaw and @NousResearch hermes; the only solution for AI to be truly useful. Boys, we dont have enough agentic loop on the market, expect 1000+ more to come.

• 
  SpaceLlama (@SpaceLlamaNFT) reported 13 minutes ago

  @parisiipunk @proroketh Same as inscriptions on Bitcoin, it's not possible to know who owns what without offchain rules. With ethscriptions you write a bunch of stuff on a transaction that appears as nonsense to an eth node alone. But some guy says hey go to my GitHub and download the software I built and we will all pretend we attached data to units of ether when we write certain specific things. The indexer is offchain, centralized, and a single point of failure. Exactly what onchain smart contracts are designed to avoid. This specific issue is even mentioned in the white paper.

• 
  Gradient Sky (@gradient_sky) reported 18 minutes ago

  @dexhorthy Define “this”. If this == GitHub issue, then you Klankär can do 20k commits a day doing these till the heat death of the universe. If this == software with scientific component, non-trivial infra and UI… well…. yeah, nobody figured that out.

• 
  TechEdgeDaily (@techedgedaily) reported 26 minutes ago

  CI breaks at 2am. Nobody notices until the morning standup. Someone spends an hour debugging. Pushes a fix. Breaks something else. Repeat. Cursor just automated that entire loop. Agent monitors GitHub, finds the root cause, opens a PR with the fix. Always on. No human needed until the review step. The AI coding tool war is no longer about who writes the best code. It is about who keeps the code working after everyone goes home.

• 
  Nate Brown (@ntbrown01) reported 26 minutes ago

  @cursor_ai I don’t want to use GitHub. I want to use my locally hosted Gitea server. GitHub isn’t really that attractive.

• 
  Thomas G. Lopes (@thomasglopes) reported 29 minutes ago

  GitHub continues to be broken? Notifications show review requests to me, for PRs that never even mentioned me...

• 
  Zeez (@wootwoot43) reported 31 minutes ago

  @AnthropicAI I cant upgrade my pro plan to max, the payment has been failing on 3 different cards. This issue had been already been raised on github and your reddit. FIx this.

• 
  Shobhit - Building SuperCmd (@nullbytes00) reported 33 minutes ago

  @mufasaland Do you mean file search in root launcher? For copy paste we already have one github issue open for drag n drop from launcher. Will prioritise that

• 
  Suryansh Tiwari (@Suryanshti777) reported 35 minutes ago

  Most AI coding tools today have one terrifying flaw: They can touch production systems directly. One wrong prompt. One hallucinated command. One overconfident agent. And suddenly your database, GitHub repo, Stripe account, or filesystem is gone. That’s exactly the problem JanuScope is solving. The idea is insanely smart: Instead of trusting the AI… put a security + policy layer between the AI and your MCP tools. So before Claude Code, Cursor, Copilot, or Codex can do anything sensitive, JanuScope intercepts the request and decides: should this tool even exist? should this SQL query be allowed? should sensitive data be redacted? should this action be audited? should the AI even see this schema? And the craziest part? It works with existing MCP servers using just one YAML file. No rewriting servers. No hosted gateway. No changing your stack. Just wrap your MCP with: npx -y januscope --config postgres-crystaldba and suddenly your AI tools become dramatically safer. The repo is packed with things that actually matter in production: • SQL mutation blocking • PII redaction • audit logs • schema injection • GitHub safety layers • filesystem protections • rate limiting • tool quarantining • OpenTelemetry support • first-run fingerprint approval And unlike most “AI security” projects… this one has real benchmarks. Across multi-question sessions it achieved: 84% fewer tool calls 84% fewer tokens ~3× faster responses because the schema gets injected directly into tool descriptions instead of forcing the model to repeatedly “discover” the database. That’s a genuinely clever optimization. But the biggest signal for me is this: The README doesn’t feel like hype. It feels like someone who deeply understands how AI agents actually fail in real-world systems. The sections on: Replit wiping databases Cursor deleting production infra SQL bypass edge cases MCP threat surfaces prompt-injection through tool descriptions …show a level of engineering paranoia that AI tooling desperately needs right now. This is one of the best MCP infrastructure repos I’ve seen recently. If AI agents are going to touch real systems, projects like this become mandatory. (Link in comments)

• 
  Connor Paton (@connorpaton) reported 36 minutes ago

  call me github the way I have performance issues