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:

• 
  TECHEPAGES (@techepages) reported 1 minute ago

  🎣 "GitBait" phishing campaign uses GitHub Pages & Google Sheets to steal banking credentials from 12+ Mexican financial institutions; no server infrastructure required 🔹 Fake bank pages hosted free on GitHub, stolen data piped straight to Google Sheets via SheetBest 🔹 100+ GitHub domains found; victims likely lured via WhatsApp, Telegram & SMS links with bank-branded previews 🔹 Active for ~3 years with ongoing development (66+ commits on one repo alone)

• 
  Pipeshub ( Open Source Alternative To Glean ) (@PipesHub) reported 5 minutes ago

  Pipelines are built. Context is broken. MCP is quickly becoming the default interface for enterprise AI agents. And that’s a good thing. It gives agents a standard way to connect with tools and data. Connecting an AI agent to Slack, Jira, GitHub, and Salesforce doesn’t mean it suddenly understands your business. It just means it can access your data silos. In short: "MCP gives your agent a passport. It doesn't give them a map." As enterprise AI undergoes a massive platform shift from passive chatbots to autonomous agentic workflows, this naive, runtime "federated search" approach creates an ugly cycle in production: - The Latency Spike: Slower agent execution while waiting for multiple external APIs to respond before it can even begin reasoning. - The Token Bleed: Skyrocketing bills from shoveling raw, unranked JSON dumps into a massive context window, praying the model finds the answer. - The Governance Nightmare: A massive risk of data leaks if you rely on a base LLM to magically guess and police complex enterprise security permissions on the fly. Agents do not fail because they lack intelligence. They fail because they lack the right enterprise context. The hardest problem in enterprise AI isn't connecting to systems. MCP solved that. The hardest problem is Context Engineering. MCP is the perfect interface, but a permission-aware context layer must be the foundation. 🚀 If AI is becoming core enterprise infrastructure, you cannot allow the strategic intelligence layer of your company to sit inside someone else's managed, closed-box platform. That is exactly why we built Pipeshub (open-source developer owned context infrastructure layer). TL;DR MCP gives agents access. A context layer gives them understanding. And deep understanding is the only way enterprise AI moves from a cool demo to secure, reliable production. 👉 Next Up Tomorrow: MCP Token Tax

• 
  Conglomerate (@0xconglomerate) reported 7 minutes ago

  Why exactly do VLAs fail? VLAs start w/ LLMs as their brain. Early roboticists (2021-2022) noticed that LLMs trained on internet text had absorbed a large amount of implicit knowledge about the physical world. So they took that best available pretrained brain, observed that actions could be formatted like language tokens, and assumed the transfer would work. But world knowledge encoded in language ≠ physics simulation. There's essentially a data structure mismatch: ▸ LLM pretraining data is discrete, symbolic, and sequential (text). ▸ Physical control is continuous, high-dimensional, and requires split-second feedback. --- ➦ VLAs in the real world, by the numbers: ① They barely work ▸ VLAs start at ~30% success on real robot tasks, it need hundreds of human interventions just to reach ~90% ▸ Best pretrained VLA hit 27.4% task progress on real robots ② VLAs can't generalize outside training ▸ On actions it's never seen, best VLAs score 25-32% task progress (fails when you change the environment) ③ Fine-tuning doesn't help ▸ The more robot-specific, the dumber it gets at everything else (only works on clean, controlled, success-only demos) ④ Too slow for a real robot ▸ OpenVLA runs at 3-5 Hz (physical control needs orders of magnitude faster than that) --- The easiest way to understand how VLAs are actually wrong is thru a real life example. ➦ Let's say you hired a chef who learned everything about cooking by reading, but has never stepped in a kitchen. If you ask them how to cook a steak, they'll tell you the best answer. But if you actually ask them to cook, they'll struggle when you hand them the pan. They'll have a hard time picking up the ingredients. They'll burn the steak. They know everything about cooking, but can't actually cook. --- ➦ Thoughts I want to take back a line I've said before: "Robots can see, but they still can't listen." (referencing to my Silencio piece before) I take it back. Robots can see, listen, even reason now. What they can't do is act in the real world. It's basically an AI chatbot wrapped in a robot body, not a robot that can actually do tasks. No wonder most demos online are scripted. There's a real problem with the brain, and roboticists have been building on the wrong foundation. VLAs are like a trojan horse, they look like the answer but bring a bunch of problems in with them. VLAs only learn through imitation which brings up the data problem. "Enough data" at scale doesn't mean hundreds of demos total. It means hundreds per task, per robot body, per environment. Hundreds again every time any one of those changes. So you've basically got a human-labor bottleneck. To get that data, someone has to physically collect it, either through: ▸ Teleoperation (slow, expensive, needs trained operators) ▸ Kinesthetic teaching (tedious, doesn't scale to complex tasks) ▸ Motion capture (high precision but high setup cost) ▸ Simulation (robots trained in sim often fail in the real world because physics engines aren't accurate enough) And you'd think, okay, maybe someday a company figures out a better way to collect all this. But the problem doesn't stop once you already have the data... Switch to a new robot body and you're collecting data from scratch, because VLAs don't transfer well across embodiments. Move it to a new environment and you're collecting again, since it just overfits to whatever setup it trained on. Give it a new task and yep, collect again, because it can't generalize to actions it hasn't seen. And if you fine-tune it for one thing, you'll probably break another, so now you're collecting data again just to fix what broke. So what was @DrJimFan and @nvidia's answer to this? World Action Models. Instead of building on a language model, you build on a world model: a model that's learned to simulate how the physical world actually behaves. VLA: a language model that learned to output actions WAM: a world simulator that learned to output actions So when you give a VLA a new task, it needs hundreds of demos to learn it. Give a WAM the same task and it simulates it forward first, acts based on that simulation, then adapts with barely any data. This is what NVIDIA did with the first WAM: DreamZero. DreamZero learns by watching the world (any video of anything, not just robot demos). The backbone is a video diffusion model, the same kind of model that generates realistic video. It was pretrained on massive amounts of internet video, so it already learned how the physical world works: how objects fall, how surfaces interact, how motion flows. Doesn't sound like an entirely different approach, right? But NVIDIA looked at it from a different angle. They figured motor actions are shaped a lot like pixels; both are high-dimensional continuous signals. So DreamZero processes them in the same model, at the same time. It predicts the next video frame and the next action together, through the same architecture. So when a robot runs DreamZero, it's literally dreaming a few seconds into the future in video, then reading its own dream to decide what to do next. If the dream looks coherent, the action works. If the dream hallucinates, the action fails. The DreamZero paper dropped last February 2026, and it's been open source on GitHub for anyone to try. Then in March 2026, at GTC, NVIDIA previewed GR00T N2, the direct successor to DreamZero. This is the production version of the WAM architecture, built for humanoid robots at scale And so far, everything's looking promising. GR00T N2 hits a 98% success rate on unseen domestic objects, a 40% jump over GR00T N1 (the VLA), and 2x better generalization than the leading VLAs. NVIDIA swapped robotics' data problem for a compute problem. Instead of collecting more human demos, just simulate more. So yeah, feels like we're finally pointed in the right direction, closer to robots that can actually function in the real world. Excited to see where DreamZero / GR00T N2 goes from here.

• 
  Anjula Dwivedi (@HeyAnjula) reported 12 minutes ago

  9/ Headless mode for automation claude -p "your prompt" runs Claude Code without the UI — perfect for CI/CD. Auto-fix lint errors on every push. Triage new GitHub issues. Generate release notes. Claude Code isn't just a tool you talk to. It's a tool your pipeline talks to.

• 
  0xSero (@0xSero) reported 16 minutes ago

  @naturevrm Dcp 4 should fix it im running it but I might need to update the GitHub

• 
  Boyuan (Nemo) Chen (@boyuan_chen) reported 19 minutes ago

  GitHub search is now an agent attack surface. A public malware-finder repo lists 9,330 suspicious GitHub repositories detected through push-pattern heuristics. Even if only a slice is ever encountered by real users, the agent failure mode is obvious. A coding agent asked to "find a library and make it work" can browse faster than it can judge provenance. Fresh commits, plausible README text, and repo-shaped packaging become inputs to an automated install path. The fix is boring and product-level: repo-age checks, provenance scoring, blocked arbitrary ZIP downloads, sandboxed installs, dependency allowlists, and logs that show exactly what code the agent trusted. For agent systems, retrieval belongs inside the security boundary.

• 
  Arti | AI Builder (@Artur_roses) reported 19 minutes ago

  Claude Code takes a GitHub issue and returns a tested, reviewed PR. No human in the loop. The new dev skill isn't writing code — it's writing issues precise enough that the agent ships what you actually wanted.

• 
  top10.dev (@Top10_Dev) reported 22 minutes ago

  SunJaycy/GoldenEye-Recomp just hit @github Trending at 503★ — the N64Recomp toolchain (the one behind Zelda 64: Recompiled / Majora's Mask) now eats Rare's 1997 engine. Static recomp ≠ emulation. The ROM is lifted to C at build time, compiled to native x86_64/ARM64, and paired with RT64 for path-traced lighting at 4K. No interpreter loop. Real binary. GoldenEye was the hard target — microcode-heavy muzzle flashes, split-screen viewport math, infamous AI. If it works, the toolchain has cleared the "Zelda-shaped problem" bar. #opensource #gamedev

• 
  Arti | AI Builder (@Artur_roses) reported 24 minutes ago

  Claude Code can take a GitHub issue, write the code, run tests, and open a reviewed PR — no human keystrokes required. The dev loop isn't getting faster. It's being removed.

• 
  Nav Toor (@heynavtoor) reported 26 minutes ago

  There is a GitHub repo that defeats Google's Play Integrity check. 61,030 stars. GPL licensed. Pushed eight days ago. The repo is called Magisk. It roots your Android phone. It hides root from banking apps. It runs Netflix on a phone the Play Store says is uncertified. It passes the same fraud detection Google built to stop it. Here is the part that makes no sense. The man who built it is John Wu. He has been maintaining Magisk for nine years. Since November 2023 he has been a Senior Software Engineer at Google. On the Android Platform Security team. The exact team that builds Play Integrity. Google hired the person who defeats their root detection. He still ships the tool that defeats it. The repo is still online. It has not been taken down. For nine years. Do not install it. Your phone is supposed to belong to Google. (Link in the comments)

• 
  Elvis Irhaye (@viii_fn) reported 31 minutes ago

  Is GitHub down or it’s just MTN trying to ruin my career?

• 
  Jesse (@jessearmand) reported 31 minutes ago

  I no longer remember why many companies started using gitlab before it went public when GitHub wasn’t owned by Microsoft. If we visit the majority of companies most tooling or software are top down driven. Only companies who build developer tools have a different mindset

• 
  Branko (@brankopetric00) reported 32 minutes ago

  AI agents are about to do to your infra what they just did to GitHub. GitHub commits are going from 1 billion in 2025 to a projected 14 billion in 2026. Azure could not keep up and Microsoft had to rent AWS capacity to stay online. That is not a GitHub problem. That is what agentic traffic looks like. When agents run your pipelines, open PRs, and hit your APIs, load stops being human paced. It becomes constant, spiky, and unpredictable. The patterns you sized your infra around no longer apply. If a 14x year broke one of the biggest clouds on earth, your capacity plan is already out of date.

• 
  YNWA🐦‍🔥 (@YNWAcrypto) reported 32 minutes ago

  The problem isn’t subtle. GitHub Sponsors has paid out ~$50M total since 2019. core-js: 9 billion downloads, running on half the top 10k sites on earth. Its maintainer was making ~$600/month when he called open source “fundamentally broken.”

• 
  李新宝 (@lixinbao_X) reported 32 minutes ago

  Just watched KK's technique. Damn. Absolute game-changer. Install 7 skills in Codex. Writing, images, covers, PPTs. Full pipeline, done. The principle is dead simple. Break the workflow into 7 parts. One skill per part. Only do one thing. Step 1 Open GitHub, find a repo. Copy the link locally. Create a project folder to save it. Step 2 Write the skill description. Input three things. What it does. What the input is. Output and acceptance criteria. Step 3 Run it and find the bottlenecks. Where it stalls Create a new skill and break it down. Don't let one skill Do 7 things it's bad at. This works for writers, Xiaohongshu creators, WeChat pub runners, Video script writers. How many skills you got installed? Have you tried it yet?