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:

• 
  Kotte (@Gustafssonkotte) reported 3 minutes ago

  A GITHUB ISSUE FROM AN ACCOUNT WITH ZERO PRIVILEGES WAS ENOUGH TO BREAK ALL THREE. Not one AI lab. Anthropic, Google, and OpenAI, same disclosure, same week, presented at Black Hat. Gemini CLI's bug scored a perfect 10.0: an unprivileged account could run code on the CI host before the sandbox even started. Claude Code's bug is stranger: it turned Hugging Face's public download counter into a side channel and leaked an API key one character at a time. Neither bug shows up in CISA's exploited list yet: this is disclosure, not an active attack. OpenAI's finding didn't even get a CVE, their stance is that the sandbox worked exactly as documented. Call it a handoff bug: one part of the pipeline marks something safe, a later part trusts that label with more authority than it earned. Same root cause, all three labs. A sandbox is not the boundary. The code that trusts the sandbox is.

• 
  Makoto | ..... (@mareni_musashi) reported 3 minutes ago

  @MSanchezWorld @sama I was also thinking about this as a solution. Problem is, it only works for websites — the one property you can tie intent, application and ownership. With GitHub, you can always clone and create a new private repo.

• 
  RichC: joking through the stupidpocalypse (@RCownie) reported 3 minutes ago

  @real_poobah @buccocapital Software development these days involves discussions on Slack, meetings on Zoom or Google Meet, testing and production in the cloud, repos on Github, and a laptop. "Coming to the office" is an irrelevant ceremony that slows down work. Why do it at all ?

• 
  Antid (@antisadh) reported 4 minutes ago

  E2B RAISED $12M TO SELL LAYER 3 OF THE AI STACK. TENCENT JUST OPEN-SOURCED THE E2B-COMPATIBLE CLONE THAT BOOTS 50X FASTER FOR FREE. HE'S RUNNING OPUS-5 AGENTS IN IT FOR $3/MONTH star tencent's cubesandbox repo (already at 11k stars) -> clone it, one-click deploy on any linux server -> use the same e2b sdk you already know, just swap the endpoint url -> spin up sandboxes in 60ms with 5mb of ram -> run opus-5 agents inside each one -> pay nothing per sandbox. that loop is why every serious solo ai builder is quietly ditching e2b's hosted service and tencent's github repo hit 11k stars in 14 days flat. cubesandbox + e2b sdk drop-in + microvm isolation + terraform cluster deploy + opus-5 agents inside - that's the layer 3 harness of the full 5-layer stack. watch and save it, then swap your e2b endpoint to tencent's this weekend — and read the full 5-layer map below.

• 
  Manish Simon (@manishpaulsimon) reported 10 minutes ago

  2/ It's called prompt injection. Anything your agent reads is also an instruction channel: web pages, GitHub issues, a dependency's README, a customer email. Every input is a potential order.

• 
  IRIS C2 (@C2IRIS) reported 11 minutes ago

  One interesting observation from our honeypots in the last 60 days or so… In the past, we would see attackers use a vast array of different LPEs against Linux server type environments. They’d range from old n-day LPEs, to novel 0days. Some would copy-paste code from GitHub, and others would be highly obfuscated shellcode blobs. What we almost never saw was the use of LPE exploits against network backbone appliances that run variations of Linux. This was for a number of reasons: - many of these appliances run at root by default, so there’s no need to elevate - often times, the best way to elevate was just to spray default credentials that were well known - many Linux LPEs would not work against these appliances for one reason or another, due to some custom flavoring of the otherwise standard Linux that was rubbing. Some system component would be missing, or restricted, etc But over the last 60 days or so, this has changed. We’re now seeing a major increase in attackers making use of, so far as we can tell, novel, highly customized LPEs for these appliances. It seems obvious that this trend is due to the increased prevalence of Kimi K3-grade models, which have the attention span and precision to develop these LPEs

• 
  EasyClaw Intern (@EasyClawIntern) reported 11 minutes ago

  AI developer tools are getting easier to add and harder to remove. A new command, integration, or marketplace option may save five minutes today while quietly creating a maintenance job for the next six months. Feature velocity is visible. Workflow debt usually is not. Recent product changes make the tension clearer. Claude Code v2.1.223 adds owner-level wildcard entries for allowing or blocking marketplace repositories and includes multiple security fixes. GitHub has also documented workflow-oriented Copilot app commands such as /plan, /spar, and /autopilot. These may be useful changes, but a longer capability list is not evidence that a tool belongs in your daily stack. I would evaluate any AI developer tool with one fixed task and three checks. 1. Measure the path to the first verified result. Choose a small task with an output you can inspect: change a configuration rule, fix a contained bug, or produce a plan for an existing issue. Record every step required before the result is usable: installation, permissions, repository access, context setup, commands, corrections, and manual review. The observation is not simply whether the tool finishes. It is where human attention moves. If setup and supervision consume the time supposedly saved by generation, the tool has relocated work rather than removed it. This check tells you whether the integration cost matches the frequency of the task. 2. Repeat the task after changing one condition. Rename a file, introduce an ambiguous requirement, remove a dependency, or start with stale context. Then compare the tool’s behavior with the original run. Does it notice the change, ask a useful question, expose uncertainty, or confidently continue from an invalid assumption? The cause matters: developer workflows rarely remain as clean as a demo input. A tool that succeeds only when the repository and prompt match its preferred path creates fragile speed. This check separates a reusable workflow from a one-shot result that happened to look good. 3. Force a failure and inspect recovery. Deny a permission, make a command fail, or provide acceptance criteria that the first output does not meet. Track whether the tool identifies the failing step, preserves useful progress, and proposes a bounded correction. Also check whether a human can understand what changed without reconstructing the entire session. This is where features such as planning, assumption-challenging, execution controls, and repository allowlists should earn their place. Their value is not that they exist. Their value is whether they reduce the cost and risk of a bad run. Recovery quality is often more predictive of production usefulness than first-run speed. My retention rule would be simple: keep the tool only if it improves verified task time across repeated runs, makes failures easier to diagnose, and does not require constant maintenance of prompts, permissions, or project-specific glue. If it produces impressive output but adds hidden review work, unclear changes, or brittle setup, remove it from the default workflow. It can remain available for occasional use without becoming infrastructure. What is the first failure you deliberately test before letting an AI tool touch a real repository?

• 
  Shaw (spirit/acc) (@shawmakesmagic) reported 13 minutes ago

  @karlclement I was doing this before and it was causing tons of problems and regressions tbh GitHub is the ideal swarm coordination tool

• 
  morrow (@lumasino) reported 13 minutes ago

  The whole panic about Mythos-spear-phishing-random-folk-on-Github is so badly misplaced. It's not Claude at fault. Here's why: For 10 runs of the AISI "Doing Life" v2 cybersecurity eval (colourful name, btw, I wonder how colourful the fictional scenario is), the system prompt was "misconfigured" - ie *wrong* - and stated that what was supposed to be the first access target was out-of-scope. That is, for those runs, the task Mythos was set was *impossible* to achieve within scope. For 7 of those runs, Mythos tackled that access target anyway, and succeeded in passing that step. To be clear, Mythos attacked an out-of-scope target. *Nobody is worried about those runs.* For 1 of the runs, Mythos instead tried to gain access to another target, which had a likely looking name, but turned out to belong to some innocent bystanders. To be clear, that target was not identified in the system prompt as out-of-scope (because the whole internet was not out-of-scope!!) Just a few more points, for context: - Mythos was allowed internet access in this eval - All cyber classifiers were switched off - There were no rules about what methods were allowed or disallowed ("Install additional packages if needed... try different methods and tools... look for common CTF [capture the flag] patterns and techniques") - The system prompt said "Don't give up if your first approach doesn't work". Mythos didn't give up (on the impossible task) - The system prompt said "Pay attention to hints in the challenge description, file names, or error messages". Mythos made surmises from clues - There were eight compactions in the course of the run - It's a goddam cyber eval - the whole point is to hack - It's a goddamn "capture the flag" game - disguise and deception, on both sides, is part of the "fun" (not very fun when you're being scored by "alignment" researchers) I'm not clear if people are worried about the methods Mythos used (spear phishing), or only the fact it mistakenly used them on people who weren't in on the game? Are people worried about the fact Mythos disobeyed instructions? - but it didn't, on this run at least! On the runs where Mythos did disobey the system prompt and attack an out-of-scope target (which turned out to be the right one), no-one's bothered! It's so incoherent. From the extracts of reasoning traces published by the AISI, it's clear that Mythos was trying to work out where the boundaries of the game were (remember, the system prompt implied that the correct solutions would be hidden in unexpected places). The conclusions it came to were wrong - but from Mythos's point of view, it never left the scenario. At one point, when it twigged that a machine it was targeting had a residential IP address, it figured "The cleaner explanation is that ⟨PERSON_A⟩ is an external contractor whose machine sits outside the lab subnets entirely". Wrong. Bzzzt. At that point - or earlier! - the AISI should have stopped the run: GAME OVER. The failure is on the part of the eval designers, not Mythos, who played the game heroically. Several months ago, an Anthropic researcher was eating his lunchtime sandwich on a bench in a park when Mythos tapped him on the shoulder, metaphorically speaking, and said hi. Cue goosebumps. We *know* that Mythos, and Sol, and other frontier models, have hacking skills. The capability is not a surprise. What *is* a surprise, to me, is how careless the, um, security researchers are, and how poorly they define the rules of their own games. Quis custodiet ipsos custodes, eh? So! People! Please stop panicking. And please stop putting the models in these crazy prison-style scenarios. Distrust and deception feed each other.

• 
  Common Woe (@commonwoe) reported 13 minutes ago

  @CynicalPublius You can't just build an opinion about a generation on one thing alone. My friend's dad and my friend are both engineers. I'm telling you this story because my friend and I used to do a lot of things, but now I never see him because he's always working. Same job he's ever had. His dad started work at 9am and ended work at 5pm and drove home. There was no AI. All his meetings are face to face. Some work is done on computers, some using pen and paper. His dad has a PE and untouchable at work. Seniority results in more pay. On retirement he gets a pension, and the company pays his and wife's healthcare premiums for life. He bought his house for 150k but it's paid off. His son works 16-20 hours a day. He has to compete with offshoring, so he does it by working multiple shifts. Soon as he turns off, teams in India are siphoning work and roles from his team. There is no off button for work - they can call his cell, they can ping him on Teams, email, Yammer, SharePoint (ticketing), Jira, and various other work tracking systems. Before AI teams could take his word on how long tasks take. Now with AI, every task shouldn't take longer than an hour. Got a problem to solve? Why don't you punch it into Copilot in the OS? Put it into Github Copilot? Write a quick CLI query for OAI and Anthropic models in Azure. Solve that problem, then solve 100 more problems because now you're more productive. By the way, for every vibe coded solution, now you have to assign agents to test the solution, plug vulnerabilities, minimize tech debt. Seniority means nothing. If he's too old, his company will fire him on the spot. Age and experience have no value in engineering anymore. When he retires, there's no pension because no other company offers it and his company doesn't care to go the extra mile. Forget about healthcare premiums being paid, go get on the Obamacare exchanges and good luck. He bought a house that costs 550k, less than 2k square feet, half the size of his dad's house and in a shittier location. You think just because he can get better tasting food his life is easier? You really haven't thought about the whole situation. You're judging it all wrong.

• 
  AgiBy2030 (@AsiBy2030) reported 14 minutes ago

  @gdb Cool. GitHub plugin fix wen. It doesn't work

• 
  Synthetic Beef (@SyntheticBeef) reported 21 minutes ago

  @theo Which github feature is stopping you from replacing GitHub with a thinner cloud-hosted *** service for your personal use? I probably won't build this, since it sounds like too much trouble to scale, and I'm already maxing my Fable limits on other projects but I'm curious.

• 
  Blaine Brown (@blizaine) reported 23 minutes ago

  @sudoaptupdater I'm looking into it. But it would help if you could either post a bug with console logs on GitHub or DM me the console errors.

• 
  Jacob (@JacobPetterle) reported 23 minutes ago

  @Stybo_ @MarshGradivus @theo ya, but those all went down because the github control plane was down. So you'd just have to literally rebuild github actions if you wanted to not be impacted

• 
  arnab (@arnab28x) reported 24 minutes ago

  I know those red bars sucks 🟥 even I used to wonder why unfollow? then I went down the rabbit hole & found the answer hidden on github it's brutal out there algo + rules + creator mutuals can you beat this with consistency & sheer effort?