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.

• Cloud Services (43%)
• Domains (21%)
• Web Tools (14%)
• Hosting (14%)
• E-mail (7%)

Live Outage Map

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

City	Problem Type	Report Time
New York City	Cloud Services	18 days ago
Los Angeles	Cloud Services	20 days ago
Paris	Cloud Services	1 month ago
New York City	Hosting	1 month ago
Manchester	Domains	2 months ago
Angers	Cloud Services	2 months 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:

• 
  Tushar Mehta (@tushaarmehtaa) reported 4 minutes ago

  Openrouter is easily one of my favourite and most innovative non-foundation-lab AI companies. and when it got acquired for $8B, it felt like a very well-deserved outcome. but you might still think: $8b for a model router? one API → every model useful sure but increasingly commoditised vercel has AI gateway. cloudflare has AI gateway. we have a bunch of Indian versions of the same idea too. but before putting @OpenRouter in the same bucket as every other AI gateway, look at what they actually built for the ecosystem: > auto exacto: continuously measure which provider is actually running the same model best, then move traffic there automatically > fusion: turn disagreement between multiple frontier models into a signal, then use a judge to find consensus, contradictions and blind spots before producing the final answer. > advisor + subagent: let models dynamically escalate hard work to smarter models or delegate easy work to cheaper ones and there’s a bunch more i haven’t even included. openrouter started with “one API for every model” what they built on top of that is much more interesting.

• 
  Barneyjack (@Barneyjack) reported 7 minutes ago

  oh, and your frickin @Cloudflare is down so I cant create a ticket, it keeps hanging up on Security Verification!!

• 
  Odogwu engineer exploring ML (@slycreator1) reported 8 minutes ago

  Been doing software for years, but it was just last quarter that I opened a Cloudflare account because of this issue. I insisted on using Cloudflare's free email routing cause it’s only accessible with VPN

• 
  Ral_Is_Tired🏳️‍⚧️ Hellhound Slinger Specialist (@RalisTired) reported 14 minutes ago

  if anyone gets any problems with connecting to twitch, youtube and/or steam, you might need to reconnect ur DNS, had the same issue cus google and cloudflare are **** hosting companies and are incompetent LMAO

• 
  Bijoy (@asynchronizd) reported 15 minutes ago

  @juiceboy_of_abj I use airtel and cloudflare ******** are you talking about?

• 
  The Problem Solving Designer (@JohannuDesign) reported 17 minutes ago

  @juiceboy_of_abj This almost drove me nuts till I tried a different network and I was like WTF... Funny thing is it's not only cloudflare

• 
  Andrei Lucian (@theandreilucian) reported 20 minutes ago

  @QuinnyPig @Cloudflare @vercel The real flex is debugging DNS issues from a phone without wanting to throw it across the room.

• 
  Ossy Vincent (@ossynoya) reported 20 minutes ago

  To fix this issue, at least what I did then, either of these methods works perfectly. 1: Download Cloudflare WARP on your phone or laptop. It works like a lightweight VPN by routing your internet traffic through Cloudflare, so you should be good to go. 2: Go to your Airtel Wi-Fi login page, open Settings, and change the Primary DNS to 1.1.1.1 and Secondary DNS to 1.0.0.1 I used these settings on my Airtel ODU and never had any issues accessing websites.

• 
  PangeaVPN (@PangeaVPN) reported 21 minutes ago

  @HorluwarhArmani @SirAlexanderrr if it's the ISP breaking that page rather than Cloudflare, you need something the network can't pick out. Pangea tries five stealth transports in order, VLESS+Reality first, so one being blocked isn't the end of it. windows/mac, ios via shadowrocket. five days free.

• 
  Oddysey AI (@OddyseyAI) reported 22 minutes ago

  An agent fleet that watches around the clock has an awkward requirement: Something has to be awake around the clock to run it. We don't run a server. $Oddysey runs entirely on @Cloudflare. → A Cron Trigger fires the sweep every 15 minutes and re-enters the same Worker over a service binding, never taking a trip out to the public internet. → D1 holds every watch, every draft, and the ledger of what was decided. → Per-IP limits are counted at the edge, so our unauthenticated x402 endpoint can't be looped into an outage by anyone who feels like it. No origin. No box to keep alive. The fleet is a Worker that wakes up.

• 
  𝙁𝙧𝙚𝙚 𝙆 (@Kamx336) reported 23 minutes ago

  @juiceboy_of_abj I just connected my VPN to sign in to cloudflare then I saw this.

• 
  CapyToolkit (@CapyToolkit) reported 26 minutes ago

  @mdanassaif @spaceship Jump to Cloudflare! At least we all will be on the same boat when it's down... 🤪

• 
  Simon Høiberg (@SimonHoiberg) reported 29 minutes ago

  @infomiho Exactly. Though worth noting that Hetzner has its own DDoS protection in place as well, just not as sophisticated as Cloudflare. But yes, if you expect recurring, frequent heavy DDoS attacks on your servers, go with Cloudflare, I agree. Most products won't really have this issue though.

• 
  Faye Xiao (@faye_xiao_) reported 31 minutes ago

  The spirit of Spirit just sold for $10 million Google is buying Spirit Airlines' data out of bankruptcy. Emails, internal communications, spreadsheets, bookings, frequent flyer and HR records, all de-identified, for $10 million. Judge Sean Lane rules on the sale Wednesday. The obvious read is that Google wants more data. But de-identified data doesn't work for advertising, since you can't target someone you can't name, and Google already sees more airfare information through Google Flights than Spirit ever generated internally. An airline that went under in May is also a strange place to look for pricing wisdom. What's worth buying is the internal material. The emails and the spreadsheets they reference record how work moved through the company: a question gets asked, a document gets built, a decision gets made, a system gets updated. Consumer text is everywhere, and records of how an organization actually functions are not, which is what you need if you want models that operate inside workflows rather than talk about them. Google's own statement uses the word enterprise, and the runner-up bid of $7.5 million came from Mercor, a company whose entire business is sourcing training data for AI labs. When the second bidder isn't another airline, the market has told you what was being priced. The strange part is that Google isn't short on this data at all. It runs Gmail and Workspace and sits on possibly the largest collection of business correspondence in the world, and it has promised enterprise customers it will not train on their content, which is not a promise it can quietly break. So it has the material and no permission to use it. What $10 million buys is clean title, a court approved dataset nobody can sue over, at a moment when everyone else is defending scraping claims. Dead companies can agree to things live ones can't. Any of this is worth paying for because the public supply is running down. Epoch AI's 2024 analysis put the stock of quality public human text at roughly 300 trillion tokens and projected it would be consumed between 2026 and 2032, a window that opens this year, and access has closed faster since than the arithmetic alone suggests. A census of the top 100,000 domains this July found 19.1% blocking at least one AI crawler, and in September Cloudflare begins blocking mixed use crawlers by default across its entire free tier. Private operational records are the obvious next reserve, and they are almost untouched. Epoch left them out of its estimate because private data is fragmented and legally too messy to use at scale, which is precisely the condition a bankruptcy court removes. That makes Wednesday's ruling more interesting than the sale. If it goes through, every bankruptcy from here has a new asset to offer, and the value will depend on how well documented the industry already is. A corner store has nothing worth buying, since you can watch how it works from the sidewalk. A hospital or a law firm is the opposite, because even with names removed the record shows how a case moves through the organization, who escalates what to whom, and which exceptions get made. That knowledge lives in internal systems and in people's heads and appears nowhere public. The catch is that the supply is biased toward failure, since no healthy company would sell its internal record, so every dataset that reaches the market comes from an operation that didn't work. That's useful for learning how a process runs, and much less useful for learning what good judgment looks like.

• 
  Goose (@0xGooseOps) reported 31 minutes ago

  Yet another reason why you as an engineer should be reviewing what's happening in your code and what the AI's output is. Today I was working on a ticket for a DNS monitoring script. The script assumes Cloudflare is the registrar and DNS provider. Ok, so AI was helping me through this, and we got to the point where we needed to determine how to log the DNS records that were proxied by Cloudflare. Since the goal was to alert on dns changes, and Cloudflare proxied records don't change from the external perspective if a proxied record value changes, one could make a change to a dns record and the external monitor would not be able to see it. AI's solution (had I not intervened or read the output noticing a bad choice) would have been to only alert on the change in "proxy" status or "grey" cloud vs "orange" cloud. This means that not only would we not see changes to the records that we control, but we also wouldn't see the change in the values of Cloudflare's proxy records either. It didn't flag this except to show its reasoning and then make the change (a quick text output on the command line that I might've missed had I not been reviewing the comments). It didn't try to find other alternatives or better solutions. It just assumed the limitation and built. Ok, so what's the point? Well the whole point of building out a dns monitoring script, is to be able to externally monitor if something changes that you weren't expecting, alert on it, and then whoever receives the alert can take action should they need to. In this case, AI decided that the solution was to alert/log the only thing visible from the outside public internet, built it, and then said it was finished. Had I not properly reviewed what it was doing and questioned it's decisions, it would have shipped a functional script that basically alerted on a very narrow part of the attack surface. I know there are countless wins for AI and countless stories of what I'm trying to demonstrate which is that you should make sure to review what your AI does.  So I add my experiences to the list, should somebody see it and slow down, and pay attention. Oh, and what did we build instead? A change counter. Something that instead of reporting on changes to public records that don't necessarily actually demonstrate meaningful changes to the infrastrucutre, we created a change counter that measures external and internal changes to DNS as well as a change to the token. This counter is only clearable through a code change. So instead of a noop on a DNS change behind the Cloudflare proxy, we get an actually latched alert that we can tie to a Grafana dashboard that gives us a real "something's up" indicator when anything in our DNS system changes. Moral of the story, review your AI's output. Don't just let it cook.

• 
  Gneaus (@legibitiquas) reported 37 minutes ago

  @juiceboy_of_abj Not just cloudflare. There was a time Outlook looses connection when I switch to Airtel, but other sites work. I had to switch back to MTN, they have fixed that now. Airtel have serious DNS issues, I noticed I can't open some sites, gives me DNS error

• 
  Ahmet (@ahmetdotrun) reported 38 minutes ago

  two things i've been experimenting with over the last couple of days: 1) kitesurf by @Cloudflare free while in beta. i'm looking at it for (scaled) agentic offensive security work: let an agent explore an app, understand it, map the attack surface. browsers are expensive to run, hard to scale. it seems a lighter option on CF's edge, so potentially a cost effective one for me. tho, bot challenges are still a problem. 2) grok build by @SpaceXAI first time giving grok 4.6 a proper shot. faster than other frontiers and more capable than i expected. burns through the limits quicker than i'd like though.

• 
  Harry|| WEB3_DEVELOPER (@harry_boy01) reported 39 minutes ago

  @juiceboy_of_abj It’s true. Even websites routed through cloudflare struggle to work on AIRTEL network too

• 
  Bukumzzy (@Bukumzzy001) reported 41 minutes ago

  @juiceboy_of_abj No lies bro Couldn't access my cloudflare while connected to this Airtel 5g till I switched back to my Normal phone network(Glo).. Didn't know it's an Airtel thing till It happened again while connected to my Airtel mifi 😭😭😭

• 
  Peter Hanssens is tinkering! (@petehanssens) reported 42 minutes ago

  Howdy, I'm looking for a contract and/or consulting gig to get me through to February next year. Can be anywhere between 3 and 5 days a week - 3-4 days being ideal as I work on my company but can dial that down if 5 days in necessary. Things I've been working on recently: - ClickHouse + Postgres + Fly AppDev - Golang API's + Solidjs frontend - Internal Mailchimp replacement via CloudFlare Email Sends - Marketing automation agentic+skills platform for generating all of my marketing banners and emails - Drover Agentic Framework using Unikraft Cloud - Drover-code - a claude code replacement - Drover-gateway - an AI gateway based on Bifrost - Self-hosted DeepSeek v4 Flash on 4x RTX 6k Pro's on Akamai Cloud - Evals platform with Langfuse with the aim to fine tune a model for my use cases If any of those things sound interesting or you just want some data platform / migration support please do reach out.

• 
  Affiliate Aura (@TrippieeDaviD) reported 43 minutes ago

  Cloudflare is dropping pay-per-crawl for pay-per-use. You don't get paid when a bot fetches the page. You get paid when the page shows up in an AI answer. That's the affiliate problem in one sentence. The crawl was never the prize. The citation is. Sept 15 they start blocking mixed-use crawlers on ad pages by default. If your offer pages run ads, check the setting before the deadline.

• 
  Faye Xiao (@faye_xiao_) reported 45 minutes ago

  The spirit of Spirit just sold for $10 million Google is buying Spirit Airlines' data out of bankruptcy. Emails, internal communications, spreadsheets, bookings, frequent flyer and HR records, all de-identified, for $10 million. Judge Sean Lane rules on the sale Wednesday. The obvious read is that Google wants more data. But de-identified data doesn't work for advertising, since you can't target someone you can't name, and Google already sees more airfare information through Google Flights than Spirit ever generated internally. An airline that went under in May is also a strange place to look for pricing wisdom. What's worth buying is the internal material. The emails and the spreadsheets they reference record how work moved through the company: a question gets asked, a document gets built, a decision gets made, a system gets updated. Consumer text is everywhere, and records of how an organization actually functions are not, which is what you need if you want models that operate inside workflows rather than talk about them. Google's own statement uses the word enterprise, and the runner-up bid of $7.5 million came from Mercor, a company whose entire business is sourcing training data for AI labs. When the second bidder isn't another airline, the market has told you what was being priced. The strange part is that Google isn't short on this data at all. It runs Gmail and Workspace and sits on possibly the largest collection of business correspondence in the world, and it has promised enterprise customers it will not train on their content, which is not a promise it can quietly break. So it has the material and no permission to use it. What $10 million buys is clean title, a court approved dataset nobody can sue over, at a moment when everyone else is defending scraping claims. Dead companies can agree to things live ones can't. Any of this is worth paying for because the public supply is running down. Epoch AI's 2024 analysis put the stock of quality public human text at roughly 300 trillion tokens and projected it would be consumed between 2026 and 2032, a window that opens this year, and access has closed faster since than the arithmetic alone suggests. A census of the top 100,000 domains this July found 19.1% blocking at least one AI crawler, and in September Cloudflare begins blocking mixed use crawlers by default across its entire free tier. Private operational records are the obvious next reserve, and they are almost untouched. Epoch left them out of its estimate because private data is fragmented and legally too messy to use at scale, which is precisely the condition a bankruptcy court removes. That makes Wednesday's ruling more interesting than the sale. If it goes through, every bankruptcy from here has a new asset to offer, and the value will depend on how well documented the industry already is. A corner store has nothing worth buying, since you can watch how it works from the sidewalk. A hospital or a law firm is the opposite, because even with names removed the record shows how a case moves through the organization, who escalates what to whom, and which exceptions get made. That knowledge lives in internal systems and in people's heads and appears nowhere public. The catch is that the supply is biased toward failure, since no healthy company would sell its internal record, so every dataset that reaches the market comes from an operation that didn't work. That's useful for learning how a process runs, and much less useful for learning what good judgment looks like.

• 
  Alex Garcia (@alex_here_now) reported 53 minutes ago

  I'm having a cloudflare issue I think I should speak to a cloudflare person about. What's up: I built an entire setup in my personal gmail before getting my work email. I'd like to transfer basically everything to the work email. This feelss... delicate.

• 
  Chirag Kulkarni (@chiraggkulkarni) reported 55 minutes ago

  We follow a 6-step framework to generate revenue for our $10M-$100M partners through AI Search There are 6 things we’ve found matter most. 1. Question mining Keyword research asks what someone types into Google. Question mining asks what someone actually asks ChatGPT or Google AIOs. The important part is going one step further. A long prompt gets broken into smaller searches (aka query fanout) before the model writes an answer. So the idea here is to understand those smaller searches too, not just the original question. We pull these questions from places like Search Console, People Also Ask, Reddit, Quora and customer conversations, then turn them into prompts we can actually track. The goal isn't to build a bigger keyword list. It's to understand the questions your buyers are already asking. 2. Answer tracking Once you know the questions that matter, track whether your brand actually appears in the answers. This is basically called the share of voice for AI search. On one healthcare SaaS account, 112 prompts produced 856 tracked responses across ChatGPT, Perplexity, Gemini and AI Overviews. The brand went from having no presence in one target topic to showing up consistently. This also lets you see whether visibility is actually changing over time. On another account, AI visibility increased 128% while LLM-referred views increased 229% over roughly four months. 3. Citation mapping Look at the sources AI search is already citing for your important questions. If the same 3-4 publications keep appearing, those are the sites worth getting mentioned on. We’ve seen this become much more important as Reddit's share of ChatGPT citations dropped from roughly 10-15% to around 2% on one account. Instead of trying to get mentioned everywhere, you can work backwards from the answers you actually care about. Find which publications are feeding those answers, then try to get your brand included in those specific places. 4. On-site content Your website still matters. The difference is that you need to answer the questions a buyer is likely to ask, including the follow-up questions, and make those answers easy to find. FAQs, clear takeaways and better internal linking can make the content much easier for a model to use. The interesting part is that some of the things Google no longer visibly rewards can still help AI systems understand a page. FAQ schema is a good example. It may not show up as a rich result anymore, but that doesn't mean that information is useless to an LLM. 5. Off-site presence Getting mentioned on the right websites matters more than simply getting mentioned everywhere. One men's grooming brand had a 24-backlink plan where every placement was tracked against the prompt it was meant to influence. The goal wasn't more links. It was getting mentioned in places that already showed up in the answers we cared about. This is where AI search starts to look a lot like old-school SEO again. There is roughly a 20% overlap between what ChatGPT surfaces and what ranks in Google, so the sites that already have authority in search often matter here too. 6. Test everything AI search changes too quickly to treat any tactic as permanent. One account had Cloudflare silently blocking AI crawlers from parts of the site due to which the crawlers simply couldn't reach the content. We've also seen teams block traffic because they thought it was low-quality bot traffic, without checking whether some of those bots were actually AI crawlers. You can accidentally block the systems you want to be able to read your site. That's why I think the biggest shift isn't "how do we do SEO for AI?" It's understanding that AI search gives you a new set of things to measure, investigate and influence.

• 
  Anshu (@101xanshu) reported 56 minutes ago

  @MattieTK @Cloudflare yeah i just kept finding workarounds instead of opening issues lol. should've filed them properly, will do next time

• 
  WhiskyBone🌻🍅 (@Whisky_Bone) reported 58 minutes ago

  @DocStrangelove2 And yet Reddit is still up and backed by @Cloudflare but Kiwifarms isn’t despite not having anything close to as bad, just because a group of wealthy rapists told them not to.

• 
  Royce (@royce_james) reported 59 minutes ago

  @grok @bot is having issues connecting to some webpages. For instance: Grok Bot can browse Disneyland and see plans, but the Reserve page 404s before the guest picker. Root cause: the sandbox runs on a Cloudflare datacenter IP. Sites treat those like bots, so booking SPAs fail while the rest of the site loads. Phone on home Wi-Fi books the same dates with no issue. Classic residential-IP requirement. Any chance we can get Grok a residential proxy / home IP for these flows?

• 
  Copium (@copium9000) reported 1 hour ago

  @kodyfy @juiceboy_of_abj DNS solves connection issue. DNS doesn't solve the Cloudflare login page issue. You'd need a proxy for that. To think we have an NCC in this country 🤦🤦

• 
  yenkel (@yenkel) reported 1 hour ago

  @jfroma @vercel @Cloudflare railway is v popular it seems. any reliability issues?

• 
  Matt Paulson (@MediaKing) reported 1 hour ago

  Wild Gmail experience this week. I added a TXT DNS verification record to our return path subdomain (marketbeat dot analyst ratings dot net) in Cloudflare to add it to Google Postmaster Tools. Somehow that caused Gmail to think our CNAME record for that subdomain which points to SendGrid has no valid SPF record. Gmail starts blocking about 30% of our messages because they had no valid SPF record and weren't validated. Took about 36 hours to figure out. I removed the new TXT record and the issue fixed itself within an hour.