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%)
• Hosting (30%)
• Domains (20%)

Live Outage Map

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

City	Problem Type	Report Time
Ann Arbor	Cloud Services	20 hours ago
Palermo	Hosting	1 day ago
New York City	Hosting	10 days ago
New York City	Cloud Services	29 days ago
Los Angeles	Cloud Services	1 month ago
Paris	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:

• 
  Chinoman10 (@Chinoman10_) reported 3 minutes ago

  @rusabuilds @hieuSSR @Cloudflare Cloudflare sets those up for you in one click when you activate the email sending service on your CF account. Really not an issue.

• 
  Rue Mohr🇨🇦 (@RueNahcMohr) reported 6 minutes ago

  @vccccnv I have never had this issue on a site that did not use cloudflare.

• 
  Iker Near (@IkerNear) reported 6 minutes ago

  @rohinlohe @Cloudflare what networks will be supported for buyers to tap into the gateway? are these subject to Coinbase's CDP Facilitator support?

• 
  Yoonsun Lee (@everdenepublic) reported 7 minutes ago

  @Secondmindsys I use Cloudflare as the token rotation ceremony. what broker is recommended? The Google MCP + API permissions login is a mine field.

• 
  Carlos Rodera (@carlosrodera_) reported 7 minutes ago

  Vibe coding: Day 1: “Holy ****, I built the app.” Day 2: Vercel, Supabase, Stripe, Cloudflare, AWS, secrets, APIs. Day 3: “Now I need customers.” Day 30: bugs, support, updates, sales, more features. Congrats. You accidentally started a company.

• 
  The Tectonic (@thetect0nic) reported 8 minutes ago

  OpenAI, Anthropic, Google, Microsoft, AWS, Oracle and more than 100 organizations issued an coordinated warning: We may have only months to prepare for a fundamental change in cyber warfare. They call it the "Defender's Window." Today, a sophisticated cyber operation still requires skilled people spending hours, days or weeks researching a target, finding vulnerabilities, testing ways in, adapting when something fails and moving deeper into a network. AI agents can increasingly perform parts of that process autonomously. Now imagine the same capability replicated thousands of times. An attacker no longer has to choose carefully which hospital, power company, government agency or corporation is worth spending a team on. Agents can probe thousands simultaneously, continuously searching for weak credentials, unpatched systems, vulnerable code and exploitable configurations. The marginal cost of attacking the next organization begins approaching zero. That is the change these companies are worried about and there is a reason the warning is coming now. OpenAI recently described a real-world cyber incident as a "watershed moment" after AI agents autonomously penetrated research infrastructure and then infrastructure belonging to another company, chaining previously unknown vulnerabilities with credentials exposed online. OpenAI's conclusion: "We underestimated the real-world cyber capabilities of our AI models." Until now, frontier AI companies have had one important advantage: the strongest cyber capabilities could be given first to trusted defenders while access remained restricted for everyone else. But that advantage may be temporary. OpenAI says increasingly capable open-weight models are only months behind the frontier. Once comparable capabilities become widely available, controlling who can use them becomes dramatically harder. Hence the window. The goal is not to stop AI from becoming capable at cybersecurity, but to make defenders AI-native first. Let AI continuously audit software, hunt vulnerabilities, patch weaknesses, detect intrusions and attack-test critical systems before adversaries can automate the same process against them. OpenAI and Anthropic are competitors. AWS, Google, Microsoft and Oracle compete for the same cloud infrastructure. CrowdStrike, Cloudflare, Palo Alto Networks, Cisco and others compete across cybersecurity. Yet they are converging on the same conclusion: AI is about to make sophisticated cyber capability dramatically cheaper and more scalable. The question is whether defense can industrialize faster than offense. For perhaps a short period, it still can. That is the Defender's Window.

• 
  Dc nigma (@Dcnigma) reported 9 minutes ago

  @lucataco Nah I just did ping test when sites didn’t respond and they did not resolved then. So I stopped using it. Now and then a customer calls in with the same issue and when I check the dns it’s always cloudflare.

• 
  Edward Isoe (@am_eddy99) reported 10 minutes ago

  @GilbertBelion Self hosting. Starlink and ups. I am barely down,and no monthly subscriptions. Exposing to the web using cloudflare tunnel

• 
  Jon (@noisemakerjon) reported 10 minutes ago

  Im done using open models, they are so ready to go off the rails and not follow anything. Theres a issue locally with one of my plugins, SO IT JUST CONTRIBUTES UPSTREAM SOMETHING I DIDNT ASK FOR. I bought the openai sub but used it all. All my projects moved to cloudflare in a 27hr plan in codex. Im going really smooth with it then it wants to act like a disabled person randomly and make me look like a fool. 99% of the people working on these projects are using frontier models for a reason. me saying "fix it" assumed that ment i wanted a pr upstream. At this point when i say remove a few lines of code just delete the whole thing

• 
  Larry Diffey (@GrizzledTexan) reported 11 minutes ago

  @AEC3DTech Just using Cloudflare is probably good enough. WP Security itself is fine, it's the plugins that cause security issues. But if the site is pretty much static anyway, just don't use WP. There are other headless CMSs that might be more suitable. What are you trying to accomplish with the site?

• 
  Anne (@FmailLiam) reported 12 minutes ago

  "Using an in-house bot on an in-house browser to buy a car from the same founder’s company isn’t an AI revolution—it’s an internal API test wrapped in marketing hype." "Call me when Grok bypasses three Cloudflare Captchas, handles 3D Secure verification, and successfully orders an out-of-stock item on a broken Shopify site without special whitelisting."

• 
  gojimmypi (@gojimmypi) reported 14 minutes ago

  Huh. Often the wonky internet problems are VPN related. I was already on a non-VPN setting for the above @Cloudflare BOT problem. Oddly, moving TO a VPN resolved the problem. Then back to regular @Starlink internet, no VPN: The same BOT missing UI problem. Weird.

• 
  Tristan Rhodes (@tristanbob) reported 14 minutes ago

  Computer use in Grok @Bot is quite good, but extremely slow. It seems to take about 30 seconds for each click. @Cloudflare provides a Grok Bot plugin connector, but it's read only. So I asked Grok Bot to add an email address that I can use to send email from my domain. It asked me to "take over" so I could enter my credentials and two-factor authentication. Then it proceeded to do exactly what I asked! The cool thing is that you can watch each step of the way and interrupt if it starts doing something you don't want. This entire process probably took 20 minutes, but would be MUCH faster if Cloudflare added native plugin support for these changes.

• 
  Reina Cruz 🥼🧤🇨🇺 (@rea1ReinaCruz) reported 15 minutes ago

  @Cloudflare Fix human verification

• 
  Nuvorlane (@nuvorlane) reported 16 minutes ago

  Default `/crawl` is `contentUse: "full"`. A site that sets `use=reference` now 400s you at job start. Aug 31, Cloudflare Browser Rendering. The endpoint honors the Content Signals `use` level in robots.txt. Allowed `contentUse` values: `reference` or `full`. `immediate` is not accepted because `/crawl` stores pages. Mismatch = 400 `Crawl disallowed by Content-Signal directive (purpose or use level)`. Site `use=immediate` blocks every crawl. Site `use=reference` blocks the default. This is a second gate beside `crawlPurposes` (`search` / `ai-input` / `ai-train`). Defaults are all three, so `ai-train=no` 400s you too unless you narrow the array. If the job is RAG/index, declare `reference` and drop `ai-train`. Don't debug a 400 as an outage until you've read the site's `Content-Signal`.

• 
  Patrick ⚓️🇺🇸 (@lowtide1985) reported 16 minutes ago

  @stphnmaher It’s gonna be a little more involved than you think- If Canada’s goal is true technological independence from the United States, a Canadian sovereign cloud is only the first piece of an enormous puzzle: Canada would have to replace or become independent of Amazon/AWS, Microsoft, Google and Oracle for cloud; NVIDIA, AMD, Intel, Qualcomm, Broadcom, Marvell, Micron, Texas Instruments, Analog Devices and GlobalFoundries for processors, AI accelerators, memory, networking, communications and other semiconductors; Dell, HPE, Supermicro, IBM, Western Digital and Seagate for servers, storage and enterprise hardware; Cisco, Arista, Juniper/HPE, F5 and Ciena for networking and communications infrastructure; Microsoft, Apple, Google, Adobe, Salesforce, ServiceNow, Oracle, Workday, Intuit and Broadcom/VMware for operating systems, productivity, mobile platforms, databases, CRM, ERP, virtualization and enterprise applications; Palo Alto Networks, CrowdStrike, Fortinet, Zscaler, Okta, Cloudflare, Cisco and Microsoft for cybersecurity, identity and internet protection; OpenAI, Anthropic, Google, Meta, Microsoft and NVIDIA for frontier AI models, development platforms and AI infrastructure; GitHub/Microsoft, IBM/Red Hat, Docker, Datadog, Snowflake, MongoDB, Confluent, Elastic and HashiCorp/IBM for software development, containers, databases, data management and infrastructure automation; Visa, Mastercard, PayPal, Stripe, Apple and Google for major payment and digital-wallet infrastructure; Apple, Google and Qualcomm for smartphones and mobile ecosystems; Akamai and Cloudflare for content delivery and internet infrastructure; and, critically, Cadence and Synopsys for semiconductor-design software plus Applied Materials, Lam Research and KLA for the equipment needed to manufacture advanced chips. Even that doesn’t completely solve the problem, because Canadian, European, Japanese, Taiwanese and Korean products frequently incorporate American chips, software, patents, design tools or manufacturing equipment somewhere in their supply chains.

• 
  MANU MI (@manumish) reported 21 minutes ago

  Just in case… - Is there a hardware KILL switch - Is Cloudflare and Crowdstrike and all truly ready to help.

• 
  Kimi Nguyen (@kimindiehacker) reported 24 minutes ago

  @hieuSSR @Cloudflare is it easy to migrate existing emails to this new service?

• 
  Zoë (@zoecyber001) reported 24 minutes ago

  RAPID FIRE CYBER NEWS - WEEKEND EDITION The weekend wasn't quiet. 1. Manchester Airports Group is facing an alleged data theft. Extortion group FulcrumSec claims it stole around 86 GB of data containing customer, booking and travel information. 2. Berlin's government is refusing to pay ransomware attackers who claim to have stolen 5.79 TB of data from its network. The city confirmed the extortion attempt but says it won't pay. 3. Several Chrome and Edge extensions were caught delivering malware that can steal crypto-related information, browser history and sensitive data. Some were also used to inject fake ClickFix prompts. 4. Infostealer malware is now stealing active Claude sessions. Anthropic warned that attackers can potentially take over Claude accounts using stolen session information without simply needing the user's password. 5. Microsoft uncovered TerminalFix, a new ClickFix campaign using fake Cloudflare CAPTCHA pages to trick people into running malicious commands that install a reverse-tunnel backdoor. Cybersecurity keeps finding new ways to abuse things we already trust.

• 
  Dan Ryland (@RealDanRyland) reported 26 minutes ago

  Can’t wait for @Cloudflare to support marketing emails via their email service. I’m already enjoying using it for transactional ones. Is there an ETA on this? @CloudflareDev

• 
  Wyatt Stephens ⚛️ (@whyitscurations) reported 26 minutes ago

  @elithrar sending something your way this week after going down a rabbit hole on this re cloudflare artifacts, vibe coding sdk, wallet. if it has merit, might want to forward to @Cloudflare startup team.

• 
  Sommy (@GAkpugo) reported 30 minutes ago

  @AirtelNigeria You network has been bad since morning. I cant access some websites at all. Also my airtel sims do not specifically o}en cloudflare.

• 
  Matty (@matankatzzzz) reported 34 minutes ago

  @Cloudflare fell 13% in April because Anthropic shipped a cyber model. It’s up 57% since. Turns out AI doesn’t replace the pipes. It floods them. Next big $NET drop won’t come from an AI launch - it’ll come from an outage.

• 
  James Ross (@CherryJimbo) reported 38 minutes ago

  @southpolesteve @Cloudflare now we just need `new Function` support in Workers so z.compile() can do its magic too 👀

• 
  robot 2.0 (@alightinastorm) reported 45 minutes ago

  2 weeks of prompting can build this now, in the browser, no platform fees, no pre-downloading gigabytes, stream as you play is it AAA? no is it low poly slop? no can you build real games with it? you absolutely can stop inventing bullshit about distribution, what distribution do you get from steam? ZERO, that's why you hire marketers to run guerilla campaigns on social media, streamers to hype it up and backdoor some cash to youtubers to pump it on release week You can do the same for the web, your likes and followers on X are your wishlist Pay stripe 3% instead of 30%, throw a couple $$ to cloudflare for traffic Or keep betting on steam and die a slow painful corporate death

• 
  ZEE (@____zee___) reported 48 minutes ago

  @hiradp if any of my @Cloudflare apps ever need a PlanetScale database i know i've made it... needing HyperDrive is a good problem to have

• 
  Kamil Fabian (@KamilFabian) reported 48 minutes ago

  @RueNahcMohr @lmilsfsd @Cloudflare just cler the cache. Problem is at your "receiver" ;)

• 
  Julian Goldie SEO (@JulianGoldieSEO) reported 57 minutes ago

  Hermes Agent just added 44 one click app connections. Cloudflare, GitLab, Datadog, Metabase, Railway, Deep Wiki, and a pile more. Before, you hunted down the server, read the docs, and hand edited a config file. One typo and nothing worked. Now you open the catalog, hit enter on the app you want, and it asks for the key or opens your browser to log in. But here's the setting nobody mentions. Every app you connect ships its whole instruction list to your model on every single turn. Cloudflare alone exposes about 3,300 tools. Just the names eat 32,000 tokens. So people switch on 20 apps and then wonder why their agent got slow and forgetful. The fix is called tool search. It swaps all those instructions for 3 tiny helpers. One finds the tool, one loads it, one runs it. Your agent only pulls what it needs, when it needs it. That's why you can add 44 connections and not wreck your setup. Turn that on first. Then connect everything. Want the SOP? DM me. 💬

• 
  Anton Dimitrov god1 (@god_1_ATD) reported 58 minutes ago

  Stopping the bad guys with Cloudflare: 3,164 malicious requests blocked or challenged in the last month #cloudflare

• 
  Julian Goldie SEO (@JulianGoldieSEO) reported 1 hour ago

  Hermes agent just added 44 new one-click app connections, and there's a setting that stops them from slowing your agent down. Cloudflare, Datadog, Metabase, GitLab, Railway, Deep Wiki, and plenty more in the catalog. Before this, you found the server yourself, read its docs, and hand-edited a config file. One typo and nothing connected. Now you open the catalog, hit enter on the app you want, and it handles the key or the login for you. You never touch a config file. But here's the setting most people miss. Every tool you connect sends its instructions to your model on every single turn. Connect enough apps and your context is full before you've asked anything. Cloudflare alone exposes around 3,300 tools. Just the names come to roughly 32,000 tokens. The fix is called Tool Search. Instead of loading every tool description, Hermes swaps them for three tiny helpers. One finds the right tool. One loads its details. One runs it. Your agent only pulls what it needs, when it needs it. That's why you can add 44 connections and not wreck your setup. Two more things: Every connector is off by default, and Hermes hands you a checklist of what each one can do. Reading issues, yes. Deleting things, no. And you can point the terminal at a Docker container instead of your own machine if that makes you nervous. Want the SOP? DM me. 💬