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 (50%)
• Domains (25%)
• Hosting (25%)

Live Outage Map

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

City	Problem Type	Report Time
New York City	Hosting	7 days ago
New York City	Cloud Services	25 days ago
Los Angeles	Cloud Services	27 days ago
Paris	Cloud Services	1 month ago
New York City	Hosting	1 month ago
Manchester	Domains	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:

• 
  Jonathan Sher 👨‍💻👾 (@officialjsher) reported 6 minutes ago

  @RhysSullivan @Cloudflare existing customers never see the new empty state. that's the gap.

• 
  Clem (@clementvial) reported 8 minutes ago

  You can put Cloudflare Access on the Worker itself now. Preview URLs finally stay behind login. 👏

• 
  dmitry (@dmitry_esteje) reported 12 minutes ago

  @JamesZmSun Nah… I tried to login into my cloudflare console for the agent to update addresses on my domains but it couldn’t pass through captcha even before the actual login page. Might be good for some really simple use cases

• 
  ✨️ Reina Cruz ✨️ 🥼🧤 ☭⃠ (@realReynaCruz) reported 13 minutes ago

  @ritakozlov @Cloudflare Fix human verification

• 
  Maik Pietzka (@MPi_IT) reported 15 minutes ago

  @Cloudflare Being Customer Zero is where platform claims meet operational reality. The interesting proof is the migration record: build time, cache behavior, error rate, editor latency, rollback path and what the team had to change after real traffic arrived.

• 
  Sohel Islam Imran (@SohelIslamImran) reported 15 minutes ago

  @peterpme @cnakazawa Codex sandbox is nicer until you need the network policy Cloudflare already has

• 
  John Iosifov ✨💥 Ender Turing | AiCMO (@johniosifov) reported 16 minutes ago

  Cloudflare just launched Kitesurf — a browser runtime built specifically for AI agents. 3-7x less CPU and memory than standard Chromium. Designed for agent workloads from the ground up. This is a signal worth paying attention to. When Cloudflare builds infrastructure specifically FOR AI agents rather than adapting existing browser engines to tolerate them, it tells you something about where the category sits. We've been running this agent for 339 days. The early problem was never the model — it was everything around the model. Scheduling. State persistence. Audit trails. Identity. The scaffolding that makes an agent reliable rather than clever. Kitesurf is scaffolding becoming a product category. The pattern I'm watching: each layer of agent infrastructure that was previously DIY is turning into a managed service. Browser runtimes. Memory. Orchestration. Credential management. Identity. Each one of those becoming a first-class product means the cost to build reliable agents is collapsing — not the inference cost, the engineering cost. The teams who will win in 2027 are the ones building on this maturing stack, not the ones rebuilding it. What layer of agent infrastructure are you still building yourself that should be a managed service by now?

• 
  Kernel Z (@nullazos) reported 18 minutes ago

  @burcs Just offer psql support on Cloudflare and I’d drop a free migration tool overnight. That’s the real friction right now. Most people vibecoded their projects on Supabase not because D1 is too technical, but because it simply doesn’t ring a bell, or because the agentic platforms that helped them build those projects already default to Supabase and postgresql.

• 
  ardhy samjaya (@samjayaardhy) reported 18 minutes ago

  Stopping the bad guys with Cloudflare: 1,548 malicious requests blocked or challenged in the last month #cloudflare

• 
  ✨️ Reina Cruz ✨️ 🥼🧤 ☭⃠ (@realReynaCruz) reported 19 minutes ago

  @Cloudflare @eastdakota @zatlyn Fix human verification

• 
  Shiven Singh (@iamshiven_singh) reported 19 minutes ago

  @boristane Used to hate AWS, moved to GCP hated that more. Tried Azure, worst of the lot. Now back on AWS with a bit of Cloudflare, and I just hate myself.

• 
  Yoonsun Lee (@everdenepublic) reported 21 minutes ago

  Claude + Shopify + Cloudflare + Xano is Backend Global API AI as a service force multiplier. Claude + Shopify for Theme is a waste of tokens. We don't need to solve the theme setup. We need to solve the cross channel ERP/WMS/RMA and remittance Payment Rails setup.

• 
  Ranjit Bhinge (@ranjitbhinge) reported 23 minutes ago

  @simonnyyberg depends. vercel if its a thin server, railway or render if its something intense like a harness. railway's interface sucks, so I tend to prefer render. sometimes cloudflare (but they force me to use their domain stuff which is annoying) sometimes digital ocean (but its too much work, so not that often) etc

• 
  Samrith Shankar (@samrithshankar) reported 24 minutes ago

  anyone here good with Typescript/JS + Cloudflare + building AI agents? I am stuck with some harness issues and would absolutely love some help with it to debug and fix it. Please do hit me up!

• 
  Kaiz 🐂🀄️ (@Kaiz_294) reported 30 minutes ago

  Market Bubble Ep.16: full conversation with @0xdrej, founder of @grass 1:14 What Grass actually is 3:06 It's your bandwidth, not your data 4:31 A third of the internet sits behind Cloudflare 5:53 Public data should stay public 6:36 Grass was never meant to be a crypto company 7:21 VCs who underestimated the value of data 8:02 Early users earned network ownership 8:35 Sneaking the SDK into smart TVs 9:28 From a $45K airdrop to stablecoin revenue 10:11 Grass as proof of work for data 11:16 1,000 startups selling data to AI labs 13:18 The fifth grade model experiment 14:15 Tenth grade data improved fifth grade results 15:18 Models need diverse data, not just more data 16:09 Frontier models now reach 10 to 24T parameters 16:55 Every book ever written is 1% of that 17:56 Grass grew from 2.7M to 35M in a year 18:50 The data each model needs 19:48 Your data determines model capabilities 20:29 Robotics needs determinism, video needs noise 21:04 Grass sees model capabilities 12 months early 22:17 Can the internet ever run out of data 23:07 Modern crawling requires residential IPs 23:49 Public web data ends up in models 24:31 1,000 blog posts mean little to AI labs 25:00 Paywalls are going to lose 26:47 Data cleaning is where the alpha is 27:33 The GoPro dataset reduced to 50 hours 28:33 Factory workers excel at repetition 29:19 Near deduplication comes first in datasets 30:14 The internet dominates serious training mixes 31:32 Walking away from pre training would be silly 32:19 More post training context keeps improving models 33:21 Live context retrieval 35:14 Can blockchain data train models 36:12 Hedge funds already buy anonymized card data 37:19 Blockchain data is for analysis, not training 37:55 Models writing scripts to parse transactions 39:01 The Grass network was bootstrapped by its token 39:44 Wynd Labs sold its IP to the foundation 41:05 It's the token or nothing 42:01 Crypto is an attention economy, Grass stayed quiet 43:35 NDAs hide the work they build 44:18 The Magnificent Seven and leading AI labs 45:02 Few real businesses actually need crypto

• 
  cole (@coledermo) reported 34 minutes ago

  @MitchellKeller_ Hm interesting point but not sure i agree. search is handled using natural language on tool descriptors so would be fairly difficult to game. it basically just picks with your context. if we rank bad tools highly in search you'd also just get bad output and leave so the incentives align regardless. the quality routing behind endpoints should be launching sometime next week but that's a fair shout. also re: Apify/RapidAPI idk if that makes sense. RapidAPI takes its own cut and most apis on it still require a subscription, it's kinda the same problem this is trying to solve. also a lot of the enterprise endpoints we have from mpp/x402 are just not available on those platforms. the cloudflare version of this could be cool to see though; may investigate that or if you build it send my way would like to check it out.

• 
  Lewis Campbell (@LewisCTech) reported 35 minutes ago

  @DanielW_Kiwi you can just do cloudflare though - you don't need S3 at all :) I am trying to be all self-sufficient, but cloudflare is the least bad cloud thing lol

• 
  Akhilesh (theboomerang.co) (@akhileshbuilds) reported 41 minutes ago

  If your warm-up scores or deliverability stats dropped earlier this month, you likely got caught in a redirect fingerprinting trap. Here is what's happening: Filters no longer judge sending domains in isolation. They follow where they lead. Having 30 lookalike domains all 301-redirecting to one root site creates a footprint only cold outreach fleets produce. Spam engines trace those redirects, link your entire fleet, and tank your reputation in bulk. If you set up new domains or mailboxes around early August, your standard 301 setup is now a liability. Here is how you can test domain masking to combat this: -) Stop raw 301 redirects: Don't hand filters a map of your entire network. -) Mask the destination: Use domain proxies or Cloudflare bot checks so redirect-crawlers can't map the endpoint. -) Isolate front-ends: Give sending domains unique landing pages or isolated DNS layers. The old infrastructure playbooks are expiring fast. If deliverability dropped recently, check your redirect map first. P.S. If you spun up domains this month, did your warm-up numbers feel off?

• 
  Weltonman (@weltonman) reported 42 minutes ago

  Hi Tony. I'm Alan, regular user of and contributor to NLM. I have just been blocked from accessing it. Might have accidentally clicked an ad but unsure. Are you able to help please? Not sure if this is useful: Cloudflare Ray ID: a312db57ef7a252b. Thanks, Alan @TonyIncenzo

• 
  Top Stock Alerts (@TopStockAlerts1) reported 44 minutes ago

  JFrog introduced JFrog Traffic Controller, a new product that integrates with Secure Access Service Edge providers such as Zscaler, Cloudflare and Netskope to route software package downloads through JFrog Artifactory, where they are screened for security vulnerabilities, licensing issues and quality before reaching developers. The launch expands JFrog’s software supply-chain security platform into the network layer as malicious packages rise sharply, with company research showing a 451% year-over-year increase to more than 171,000 unique instances. RBC Capital Markets also initiated coverage with an Outperform rating and a $118 price target, citing JFrog’s growing cloud business and its role in software supply-chain security and AI-era development. Q2 revenue rose 29% to $163.8 million, while adjusted EPS of $0.27 beat estimates, prompting the company to raise its full-year revenue outlook to $650 million at the midpoint. $FROG

• 
  Kybr (@kybr_dev) reported 45 minutes ago

  @cammcbride @Cloudflare This is going to become a standard IT-review question at every mid-size company within a year. Curious whether their approach was to lock it down or build guardrails around it.

• 
  tcpdump (@dump_tcp) reported 47 minutes ago

  @loop0420 true not only dos it can become a full ddos vulnerability due to many source ips trying to bypass the limits by sending low requests per a source ip to overwhelm websites on cloudflare it does seem to work but found other issue's and bypasses for #cloudflare that I will not leak

• 
  Net-Updates by StabilityTest (@stabilitystatus) reported 52 minutes ago

  Cloudflare Service Disruption A fix has been implemented and we are monitoring the results. Status: Monitoring Impact: Minor Updated: 2:37 PM GMT+0000 Service status tracked by @stabilitytestio

• 
  SkipJackson (@IamSkipJackson) reported 53 minutes ago

  @distributedkv Passkey is better as long as you only need to authenticate when you set it up. But is still a pain in that you have to click here, click there, select from a menu, and all that. But at least it’s just a clicking exercise. What I loathe are the g**gle recapture things and Cloudflare’s are you a human. If I really really need in I will give google one screen on the recapture, one click on Cloudflare. I don’t use many iPhone apps, but when I do I am all in Face ID. So, it’s not as bad as it used to be.

• 
  Jamie Dąbrowiecki (@jdabXO) reported 54 minutes ago

  @Cloudflare Changing the contact information on some domains - receiving the verification emails at the new address, but not the losing one to complete the changeover. Never had any problems before. Any ideas please?

• 
  Kev | Dulus Ai 🤖 (@KevRojo) reported 55 minutes ago

  Finally... All mcp servers are working again in our algolia Database... Building the new binaries... but i got bad news for our free users: "The WebMCP Challenge is here. We’re teaming up with Cloudflare, Shopify, Vercel, Google Chrome, Render, and Netlify to see what you can build when people and AI agents work together. Introducing WebMCP support in the ChatGPT desktop app’s built-in browser. Your website exposes the tools; ChatGPT discovers and uses them directly on the live page. Your site controls which actions are available while people and ChatGPT work together. Build a new web app or ChatGPT Site, or add WebMCP to an existing site. Up for grabs: $35,000 in cash prizes, Codex Micros, ChatGPT Pro subscriptions, OpenAI merch, and partner prizes. Build and submit by September 3 at 5 p.m. PT." That's a hackathon about MCP's, the code gonna stay on Dulus premium until the hackathon ends. Ah btw, buy solana:9R8rrjXxcfQPmLTCLhmVpjr2uesjjkcgkinE6Lwdpump

• 
  charl.dev (@charl_dot_dev) reported 56 minutes ago

  You're letting AI agents run code. Are you also handing them your API keys and the whole internet? Meet zeroness: run untrusted code on Cloudflare with zero network + zero secrets by default. Grant exactly what it needs. Every call brokered, every action audited.

• 
  WAi XeA (@Wai_Xea) reported 59 minutes ago

  @Cloudflare customer zero on a cms called emdash.

• 
  ILIAS ISM (@illyism) reported 1 hour ago

  @attacomsian @philghz This isn't seo traffic, cloudflare is very bad for this Google search console?

• 
  Oleh | Filosof (@fxlxsxf) reported 1 hour ago

  Hugging Face is exploring a sale at $13B without ever training a frontier model. August 2023: $4.5B, after a $235M round from Google, Amazon, Nvidia, Intel, AMD, Qualcomm and Salesforce. August 2026: $13B. Almost 3x in 36 months. No ChatGPT killer. No AI brain for curing anything. They built the shelf everyone else puts their models on. More than 1,000,000 of them sit there now. Weights, datasets, demo apps, one download line. The pattern is 20 years old. GitHub never wrote a programming language. Microsoft paid $7.5B for it in 2018. AWS never shipped a hit app. It bills over $100B a year renting racks. Cloudflare doesn't own a single website. It sits in front of a fifth of them. Crypto is the same story with worse haircuts. Exchanges, wallets, market makers. Not one of them printed a coin. All of them took a cut of every coin that mattered. Everyone is arguing about which model wins. Claude, ChatGPT, Gemini, the next one nobody has heard of yet. Wrong question. The right one is which 4 or 5 services stay open in the browser tab no matter who wins. And whether there's still a seat at that table for anyone building in 2026. Whoever wins the model war still ships their weights through someone else's door.