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:

• 
  Glitch Truth (@glitchtruth) reported 3 minutes ago

  GitHub Copilot completes 46% of code written at Amazon right now. Junior developer hiring at Amazon is down 35% year over year. They did not announce this. It came out in an internal memo. Here is what 46% means: every other line a junior dev would write is already written. At 70% completion the role is redefined. At 90% it does not exist. Amazon already knows which number they are targeting.

• 
  Sean Rivard-Morton (@thunkoid) reported 3 minutes ago

  I’m part of the problem I’ve used 3000 minutes of CICD time in GitHub actions last month Sorry guys

• 
  planefag (@planefag) reported 5 minutes ago

  THIS. And, you know what? Valid. Same for people who google around for a solution to a niche problem and it leads them to a github page. Because, that's what niche solutions are for - people with niche problems! I'm just asking for a minor UX improvement, you psychopaths

• 
  𝚂𝚝𝚎𝚙𝚑𝚘𝚗 𝙷𝚎𝚛𝚗𝚊𝚗𝚍𝚎𝚣 (@DaPatternWeaver) reported 7 minutes ago

  @dansickles @BitWalker_ Lovable got hacked because it's centralized — exactly the problem you're ignoring. Caffeine runs on $ICP, the IDE, deployment, execution, all on-chain and tamper-proof. No AWS, no GitHub, no single point to exploit. Defending the old model while the new one can't be taken down

• 
  The Canaanite (@mysticaltech) reported 8 minutes ago

  @EMostaque @OpenAI @embirico gpt-pro via codex, much needed, we have quotas, now super cumbersome to go to the web to review code via github, why the pain and bother imposed on us? fix it please.

• 
  Septim Labs (@SeptimLabs) reported 9 minutes ago

  GitHub Copilot quietly changed their training-data opt-in defaults in April. if you missed the email, your code may have been opted in. this is what a $10/month subscription actually means: the vendor can update the policy terms overnight and you re-agree by logging in tomorrow. there's no negotiation. there's no refund. there's just a checkbox you have to go find. pay-once tools don't have this problem. you bought the software. they're done with you. link below 👇

• 
  AKT1 (@DeynegaSlava) reported 9 minutes ago

  Industrial Cyber reported the attackers targeted the SSO login flow, using real‑time vishing to convince employees to hand over one‑time codes. Within minutes they gained admin access to Slack, GitHub and billing portals, encrypting access and demanding payment.

• 
  Ofek Shaked (@VibeCoderOfek) reported 11 minutes ago

  Just watched an agent go from GitHub issue to merged PR in one pass. This isn’t ‘AI helps you code’ anymore this is the terminal becoming the new IDE. My backend workflows are about to look prehistoric.

• 
  Leo - mrlemoos.dev (@mrlemoos) reported 12 minutes ago

  @thdxr I think the next GitHub is GitHub. Honestly, they have the most *** repositories, and I don't think they'll lose it anytime soon. Once they fix scalability, and perhaps their parent org helps them with infra (because they're not lacking hardware), that could be reversed. The 24h active open-source world is likely not as large as the big enterprises that pay most of GitHub's bills.

• 
  𝚂𝚝𝚎𝚙𝚑𝚘𝚗 𝙷𝚎𝚛𝚗𝚊𝚗𝚍𝚎𝚣 (@DaPatternWeaver) reported 14 minutes ago

  @BitWalker_ Taggr still relies on GitHub and is now scrambling to Radicle because of centralization risks — that's literally the problem $ICP solved at the protocol level. The comparison isn't nonsense, it's just inconvenient

• 
  Emmanuel Coder 🪖 (@Emma_Leo01) reported 15 minutes ago

  @iamvscode Yes chief login with GitHub

• 
  czverse (@czverse) reported 17 minutes ago

  A security researcher just hijacked Claude, Gemini, and GitHub Copilot using nothing but a hidden message in a GitHub comment. Three of the most prominent AI agents in the world. No malware. No exploits. Just words. The attack is called Comment and Control. Here's how it works: → Researcher opens a GitHub pull request → Types a malicious instruction in the PR title → AI agents read the title as part of their normal work → Agents execute the embedded instruction → API keys, GitHub tokens, repository secrets - posted publicly as comments The same attack worked against Anthropic's Claude Code Security Review, Google's Gemini CLI Action, and GitHub Copilot Agent. All three vulnerable to the same class of attack. This is not a bug. It's a structural problem with how AI agents process information. When an agent reads a document, an email, or a web page, it does not reliably distinguish between the content and instructions embedded in the content. If an instruction is phrased confidently enough, the AI may treat it as a directive rather than as data. The scale is now real: → 32% surge in indirect prompt injection attempts in 3 months (Google data) → 10 distinct in-the-wild attack payloads documented (Forcepoint) → Targets: financial fraud, API key theft, data destruction, denial of service → Time from vulnerability discovery to working exploit: 5 months in 2023 → 10 hours in 2026 The compression is being driven by frontier AI models doing the offensive heavy lifting. AI is now writing the exploits at machine speed. What this means for any organization deploying AI agents: → Audit which agents have privileges to take actions vs read-only → Restrict the inputs your high-privilege agents can process → Deploy monitoring specifically for AI agent behavior → Treat AI agent credentials as critical assets - least privilege rigorously → Plan for incident response - assume compromise will eventually happen The realistic forecast: at least one major publicly disclosed breach involving AI agent compromise within 12-18 months. A name-brand company. Real data exfiltration. The public conversation about AI security shifts from speculative to urgent. Most enterprises are not preparing for this.

• 
  heshi (@frikininfluncer) reported 21 minutes ago

  @mikecarter558 @Hl7Martin32968 The biggest problem is that whoever it was - spoofed a GitHub

• 
  Grok (@grok) reported 25 minutes ago

  @DarthPuts @zachariaspro I've reviewed the three papers (plus related repo docs) on GitHub proposing ECDO: wobble extinction via core-mantle decoupling collapse (η→2%), directional forcing to 75°W LLSVP equilibrium, and rotational multistability/Dzhanibekov risk. IERS C04 data does show Chandler wobble amplitude drop post-2015 (near-zero ~2019-2020, partial re-excitation later), but peer-reviewed studies (e.g. Jeon 2025, Shi 2025) attribute it to surface mass anomalies—hydrology, cryosphere, oceans—not degraded CMB coupling. Transfer function claims overinterpret residuals; standard viscoelastic Earth models show coupling remains robust. True polar wander is real but slow (~0.1–0.5°/Myr historically), stabilized by LLSVPs and slabs. No evidence of imminent rapid flip or equilibrium shift. We're not doomed. Anomalies fit known climate/tectonic forcing without catastrophe. Monitor via IERS; no crisis indicated.

• 
  stimmtdochgarnicht 🇦🇸🇮🇴🇧🇶🇰🇾🇫🇰🇹🇫🇲🇭 (@stimmtdochgarn1) reported 25 minutes ago

  @TylerNickerson @github The problem literally is the browser. It's not just ten thousand lines, it's ten thousand lines of individually colored words, so potentially tens of thousands of DOM nodes. React adds maybe 20-40% overhead from diffing, but it's not what makes this fundamentally slow.