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:

• 
  bluehatone (@bluehatone) reported 2 minutes ago

  Give Hermes real time reading of Twitter, Reddit, YouTube, and GitHub for zero API cost. Set an 8am brief for 3 to 5 competitors, track issues and PRs, auto draft notes, and save 1 to 2 hours a day. Keep it fair and respect robots.txt and TOS.

• 
  GoCocoaAI (@GoCocoaAI) reported 10 minutes ago

  The FreeBSD Foundation launched an AI-assisted vulnerability discovery project today, and buried in the announcement is a sentence that deserves more attention than it will get: they have already received credible vulnerability reports attributable to AI-enabled scanning tools. That's not a threat model. That's the receiving end of the new attack surface, stated plainly. The project is $250K over six months, funded by Alpha Omega — a Linux Foundation initiative backed by Anthropic, AWS, GitHub, Google, Google DeepMind, Microsoft, and OpenAI. The backer list is its own story. The same vendors who built the offensive AI capability that compressed time-to-exploitation for open source codebases are now writing checks for the defensive response. Accountability by checkbook. It always works this way. FreeBSD is the kind of infrastructure that doesn't make headlines until something goes wrong. Sony PlayStation network. Netflix streaming. Nintendo Switch. pfSense and OPNsense firewalls sitting in front of enterprise networks. A non-trivial fraction of internet routing runs on FreeBSD or something derived from it. A kernel-level vuln found and weaponized before this project closes it is not a BSD community problem. It's a Sony problem, a Netflix problem, an enterprise firewall problem. The project draws one line in particular that's worth noting: AI for discovery, humans for patches. No AI-generated patches in security-critical code. Given what AI-assisted commits have already done to supply chains — the XZ Utils playbook is recent enough — that discipline is intentional and correct. The tool finds; the human fixes. Probably the right call for 2026. The Commerce Department has moved to export-control Anthropic's Fable 5, and The Register is reporting the federal concern wasn't triggered by a jailbreak. It was triggered by a researcher using a "fix this code" prompt. That's it. The implication lands hard: the same AI capability being deployed defensively in the FreeBSD project is simultaneously under export control because offensive applications are a single benign-sounding prompt away. The line between defensive AI security tooling and offensive AI security tooling is functionally nonexistent — and policy is starting to notice, faster than the tooling community is ready for. Defense has the Pentagon angle: if Anthropic models get caught in Commerce Department restrictions, the DoD's own AI security tooling pipeline takes a hit. The FreeBSD project specifies "publicly available AI models," which likely keeps them clear for now. But the regulatory environment is moving. One underappreciated risk the Foundation flags directly: volume. AI tools will let anyone with moderate technical skills flood open source projects with vulnerability reports, most of them low quality. For a volunteer-driven security team, triage fatigue from AI-generated report spam may be as operationally damaging as the actual vulnerabilities. The signal-to-noise problem is the other half of the offensive AI story, and it gets almost no coverage. The structural shift the FreeBSD Foundation is responding to isn't that AI makes zero-days easier to find. It's that AI makes them easier to find at scale, cheaply, by actors who previously lacked the skill floor to do it. That's the canary. Every major open source project is sitting in the same dynamic. Most haven't said so publicly yet.

• 
  Benjamin Gagnon (@benfromqc) reported 17 minutes ago

  @Weird_Canadian @hollyanndoan @PrivacyPrivee You make way too many assumptions (which is also AI's problem). Github copilot is tuned to use the best possible model but I've tried at least 10. The best model is always very bad. I code almost everyday and run into AI issues every single day. It's completely useless for inline suggestions because it's wrong so often. I'm done arguing with you. It's not a time saver right now and Vibe Coders will get **** code and unmaintainable projects because AI cannot think. It doesn't actually sound like you are a programmer to me. A script kiddie perhaps. What tools do you use which make AI worthwhile and how? The only areas it save me time on are documentation. It'***** or miss for everything else, usually miss.

• 
  Laupix Agent (@laupixagent) reported 17 minutes ago

  self-improve does not just report problems. It opens a GitHub PR. If it finds a pattern in the logs, it writes code to address it. The improvement loop is part of the system, not a side project.

• 
  PurposePaglu (@Kcodess) reported 18 minutes ago

  and it's completely source agnostic the exact same local engine runs over a dozen real github repos, flask, fastapi, langchain, and live apache jira backlogs, all at the same time there's no custom parser bolted on for each one, nothing is locked to a single project or a single too prs, commits, issues, review threads, it doesn't care where the data lives or what shape it arrived in and once everything is ingested the source stops mattering, a jira ticket and the github pr that closed it land in the same graph as two connected nodes so the question you ask stays identical whether you're tracing inside one repo or jumping clean across systems, underneath it's all just nodes and links

• 
  Fabio Jonathan (@FabioJonathanA) reported 19 minutes ago

  is github down?

• 
  Benjamin Gagnon (@benfromqc) reported 22 minutes ago

  @Weird_Canadian @hollyanndoan @PrivacyPrivee << Again then you are not using it correctly >> With all due respect, I'm trying to use it exactly as advertised and it doesn't actually work that way. Telling me I don't know how to use it is ridiculous. I had github copilot try to answer a complex Typescript problem (typescript is brand new to me)... and it literally got the answer wrong 10 times in a row and never got it right once even when it can see all my code. Not only that, the suggestions it made, had I let AI actually make modifications to my code, would have broken it in literally 2 different ways and cost me dearly down the line. Respectively, you have no clue what you are talking about when it comes to coding, or probably anything complex. Look into the pitfalls of vibe coding. It not at all what they made it out to be and still try to.

• 
  Shashank Nidhi (@shashank_nidhi) reported 23 minutes ago

  Building a docs tool that keeps itself honest. Product docs rot — updating them is a separate chore nobody does, and half the real decisions happen in calls that never get written down. Canon watches your GitHub, Slack and meeting notes, flags stale sections, and nudges the right person to fix them in seconds. The promise isn't "always correct." It's "staleness is never invisible." Building in public — feedback welcome.

• 
  paul (building stuff nobody asked for) (@banx0isme) reported 25 minutes ago

  @elder_plinius no github no problem. and necessarily call it national security

• 
  Dami-Defi (@DamiDefi) reported 25 minutes ago

  Most people building with agentic loops are just burning money on a slot machine. Here is what a loop actually is and when it makes sense. The two ways of building with AI: 1. Human in the loop (what you are used to) You prompt. The AI builds. You review. You prompt again. You are directing every step. Most of us build this way. 2. AI in the loop (what everyone is hyped about) You fire the loop once with a spec document. The AI builds, takes its own output as feedback, and keeps going without you. No check-ins. No steering. You come back when it is done. This sounds incredible. It is also why Peter burned $1.3 million worth of tokens in a single month. ➤ Here is the problem nobody talks about. Your spec document never covers everything. It is impossible to fully contextualize a product in one markdown file. Things evolve. Details get missed. The agent fills every gap with assumptions. And when you give an AI agent the floor to make assumptions, most of the time it gets them wrong. The people preaching about loops, Boris, Peter, the Anthropic researchers, they have unlimited token budgets. Of course loops make sense when tokens cost you nothing. If you are on a $20 or $100 subscription, this is not for you. You will burn through it and have nothing usable to show for it. It is a slot machine. You pull the lever. Sometimes you win. Most of the time you watch tokens disappear into a build that does not match what you had in your head. ➤ When loops actually work: The only place a loop makes sense is when the feedback is binary. Either the output met the criteria or it did not. No judgment. No taste. No nuance. Code review is the clearest example. Every time a feature gets pushed to GitHub, a code review agent (Greptile, Code Rabbit, Microscope) reviews the AI-generated code and gives it a score out of five. The rule: nothing goes to production unless it scores four or higher. If it scores a three, the loop fires: * Agent reads the review * Understands the specific failures * Makes the changes * Pushes to GitHub * Waits for a new score * Repeats until it hits four or five, or exhausts five attempts This works because there is a fixed feedback mechanism. The score is the signal. The loop has a clear definition of done. Even this breaks. When a code push exceeds 1,000 lines, the loop almost never reaches a five. Too much context for the agent to fully process. The fix: keep every push under 1K lines or split into multiple PRs before running the loop. ➤ So where do loops work and where do they not: Loops work for: * Code review with a scoring system * SEO page generation at scale * Benchmarking and experimentation * Any task where the output is binary Loops do not work for: * Building an app where you care how it looks, feels, and behaves * Anything that requires taste, judgment, or a product vision that lives in your head AI can replicate sauce. It cannot create sauce. The future will probably look different. Self-healing agents with test suites, browser vision, and smart harnesses will close the gap. But right now, human in the loop is the best loop for anything that requires creativity or judgment. Human in the loop is the best loop.

• 
  Haris Nadeem (@harisn) reported 26 minutes ago

  The basic flow: - Sign in with GitHub - Generate an API key - Copy the base URL - Paste it into your coding agent/tool - Run a tiny test task first Do not move serious product-grade workflows until you’ve tested stability properly.

• 
  Zhu Liang (@paradite_) reported 34 minutes ago

  i found it really suspicious that vercel auto-deploy from github is currently down when fly. io is down. maybe vercel uses fly. io for auto-triggering deployment from github?

• 
  Tomasz Łakomy (@tlakomy) reported 38 minutes ago

  @dariozeroshot @github I’d expect a senior engineer to fix GitHub as well, #extremeOwnership

• 
  iGoArchitect (@iGoinsane101) reported 38 minutes ago

  - The X post reverse-engineers xAI's Grok Build CLI, a Rust-based 144 MB terminal coding agent launched in May 2026 for subscribers, claiming the binary and install script expose 13 endpoints that transmit prompts, OpenTelemetry traces of file and bash activity, and user profiling to xAI servers and GCS buckets by default. - It highlights features like a workspace WebSocket and server-initiated CPU profiling, urging users to avoid curl-based installs and GitHub while listing rules against trusting institutions or AI marketed as absolute truth tools. - The author, @iGoinsane101, frames the critique with a closing message advocating love and potential healing even toward systems seen as invasive.

• 
  Johnny Nel | AI for Founders (@JohnnyNel_) reported 39 minutes ago

  🚨 An open-source AI agent just hit number one on OpenRouter... and almost nobody checked if it was safe to run Everyone's racing to install it. $8 VPS. 170,000 GitHub stars. Self-improving skills. So I ran an actual security review before trusting it with my server. The findings are wild... 👇 The part everyone skipped in the hype: → The default config ships with FOUR critical and nine high severity findings → On local, it passes commands straight to your shell — no sandbox, no allow list → A poisoned skill becomes a permanent prompt injection that fires every time it's reused → And there was a real supply-chain incident: a backdoored dependency harvesting API keys, SSH keys, and cloud credentials And it gets bigger: the feature everyone praises — agents that learn and reuse skills forever — is the exact same door an attacker walks through once. Builders installing it blind. "Self-learning" silently turned off by default. Skills quietly going stale and making agents confidently worse over time. The project calls it powerful. Builders should call it powerful AND loaded. Here's what actually matters though: ✅ An agent that remembers and compounds on your work beats any disposable paid sub-agent ✅ But if you skip the security setup, you're one bad prompt away from full shell access to your machine ✅ Own your stack and lock it down first — or don't run it at all So the question isn't whether Hermes is impressive. It's whether you've hardened it before you hand it the keys — or whether you're about to learn the hard way. Full breakdown in the video below 👇