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%)
• Hosting (21%)
• E-mail (7%)
• Web Tools (7%)

Live Outage Map

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

City	Problem Type	Report Time
New York City	Hosting	1 day ago
New York City	Cloud Services	20 days ago
Los Angeles	Cloud Services	21 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:

• 
  *yo (blue tick) (@roramora0) reported 2 minutes ago

  @Cloudflare does a neat trick to make its AI chat much harder to abuse. Too bad GLM 5.3 is much smarter.

• 
  Abomination (@Abomination81) reported 4 minutes ago

  @ohiobug981 ******* ********. Same **** happened with the co-location, same thing with market maker rebates, now they are allowing them to bypass cloudflare giving a latency advantage, now this...

• 
  scott (@scottonote) reported 4 minutes ago

  @NateMeyvis correct primitives can help mitigate this. MVC for "classic" web request/response, cloudflare Durable Objects for actor pattern, polars over pandas. something something map reduce

• 
  Welldone (@welldone_tech) reported 5 minutes ago

  Cloudflare shipped WriteGuard for MCP servers: agents can read data, write operations need an explicit gate. Unrestricted tool access is the actual production risk with agents - not that the model is wrong, but that a prompt injection or a hallucination reaches a database that never should have been writable in the first place. Read/write separation for agents isn't a Cloudflare feature. It's the same principle behind every gate we put in our own process - a senior approves the spec before code exists, a different model reviews it, an engineer deploys manually. The tool doesn't get more trust than the person watching it.

• 
  Brandon Beery (@NDominator) reported 5 minutes ago

  One of the Holy Grail problems at Press Ganey was theory crafted around happiness. It was a side comment at a retreat. Not to be taken seriously. How to survey and have predictive analytics around a happiness equation for healthcare. Can you know what will make a patient happy? There's a hundred reasons why it's a golden goose that'll never exist... But to try to do that now with LLMs as a private practice? A single doc and a dream with their $20/mo Codex subscription? Not a 30+ year old company purchased by Qualtrics for their MASSIVE database and expertise in the field? Good. Luck. Can AI help with better health outcomes? ABSOLUTELY. Can skills, agents, harnesses, frontier or even local models be a big part of that? Yes! Skills are just SOPs for bots. If you can teach a person how you do something you can teach a bot. I have a cron job through my Hermes agent that texts me every day at 11AM asking how my back is doing. My answers get scored, my notes taken down, and sent to supabase. Then I have a dashboard of the data up on a cloudflare site so I can see it anywhere on any device. But that's because it's just for me. I wanted to know what was causing my back pain and what made me feel better. I have agency over my health and health data. I can't imagine the red tape a doc would have to go through to have that I do daily setup for their patients to not get nailed into oblivion from CMS, HIPAA, etc. Just the security on the message would be a nightmare. If you want to do this, you don't teach the patient how to ruin your life 24/7 with AI. Which is what you'd get if they were messaging you with their latest daily updates. No one wants to deal with that, let alone your doctor who is used to 6 minute appointments. You do this instead: You take your specialty, what makes people go to you, and you record it. Document it out. Transcribe it. Get as much of you down digitally as possible. Then you take digital 'you' and make that skills. That's your views, your beliefs, your techniques, your skillset, the nice testimonials people have about you, and all of your case studies and stories anonymous and generalized. Take your bundle of skills and package them all up as on demand consult plugins. They never send you their data. Everyone plays on their own machine with their own risk and liability. They are glorified biohackers tinkering with their own body, not patients under your care. I say plugins because if they are using Codex or Claude Code/Cowork it'll do everything you'd do end to end with many skills. No copy/pasting between a zillion chats.. That's just confusing. It's the same decision tree you'd work through if you worked with them in person. Literally to your tone, the questions you'd ask, the logic in your brain just worked out in code with their context. But they are stepping through their own health data, their own journey just more informed by your opinion. Not health or investment advice. Anyone can do this for themselves.

• 
  Tobi (@To_bi_Bakare) reported 5 minutes ago

  It is. But Cloudflare doesn't support Naira, and most Nigerian bank cards are naira-restricted for international payments by default. A few ways around this🧵

• 
  PrimeLine (@PrimeLineAI) reported 7 minutes ago

  @zlxndr @val_riabtsev the 44-tool problem has a harness-level fix now, which i think moves the argument. tool schemas can be deferred - the server's tools get listed by name only, and the full jsonschema is fetched on demand when a task actually needs it. my session right now: 123 mcp tools listed across 5 servers, 16 schemas actually pulled. that's the cloudflare search()/execute() shape without the server having to be designed for it. so "schemas bill you every turn whether you touch them or not" is a harness default, not a property of mcp. which makes the real variable eager vs lazy loading, rather than cli vs mcp.

• 
  Dan Ryland (@RealDanRyland) reported 16 minutes ago

  @philcampbell Are you using the Resend route or Cloudflare's new email service? I'm a big Cloudflare user and fan so wondering if I could create a product inbox to take all of my various domains into one inbox from forking this

• 
  Robert Youssef (@rryssf) reported 16 minutes ago

  Claude and ChatGPT still can't do this. Cloudflare just open sourced a computer for AI agents, and it fixes one of the most annoying limits in the entire agent stack. Every time you start a new session with Claude or ChatGPT, whatever files or context the agent built up just vanishes. The agent writes notes, tracks progress, remembers where it left off, and then the session ends and all of it is gone. You start from zero again. Most agent frameworks also spin up a full Linux container just to run a simple command like sorting files or grepping through a directory. That's a lot of overhead for something this basic. Until now. Cloudflare shipped Computer, an open source, MIT licensed runtime that gives an AI agent its own persistent machine instead of squeezing every task into a throwaway container. The core idea is a virtual filesystem that lives inside a Durable Object. The Durable Object holds the real state in SQLite, and it exposes one pluggable execution surface through workspace.runtime. Because the state is durable, the agent knows exactly where it left off automatically. No manual saving, no rebuilding context from scratch. The traction is staggering: It hit over 8,000 GitHub stars and 456 forks within days of shipping, and it trended to number one on GitHub across every language, not just TypeScript. Cloudflare claims it only needs to spin up a full Linux container about 10% of the time. The rest runs directly through the durable filesystem. Claude and ChatGPT still treat every session as disposable, and every trivial file operation as worth a full container. If durable, cheap memory becomes the default for agents, the container stops being the unit of compute. The filesystem is.

• 
  Cooper (@GarudaO7) reported 16 minutes ago

  @nsptsaiphanitej yes. edo jio side dns issue anta *.up.railway.app sites ki, cloudflare through proxy cheyali or custom domain use cheyali

• 
  Oddysey AI (@OddyseyAI) reported 21 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.

• 
  Nick Blow (@NickBlow) reported 21 minutes ago

  @hackerdocc Vercel’s main value prop was making AWS easier to use but with agents they kinda solve that problem already, trivial to do some iac stuff. Fluid compute is genuinely pretty cool and not many companies compete - though vercel is still v v expensive even with the efficiency there. Cloudflare skirts the issue by charging for CPU vs wall clock. Cloud run and lambda managed instances are *sort* of analogous but not quite.

• 
  Clubs (@C1ubDeuce) reported 24 minutes ago

  This is going to sound awful but I bought a cloudflare domain like three months ago and I have no good idea on how to actually make a website with it does anyone know how to do this

• 
  Matt Smith (@msmith508) reported 25 minutes ago

  My router started having issues downloading a firehol ip block list…problem appears to be compound: they now block forcing redirect to but even after importing trusted root certs, it still fails, due to cloudflare most likely…my current workaround is dl from github

• 
  The Civic Lens (@OriginalOGDAW) reported 28 minutes ago

  @TorBox It people knew they could just use free cloudflare and make there own Debrid service for free with 10tb monthly limit

• 
  Shadooow (@ShadooowOnX) reported 29 minutes ago

  Andrej explains why Grass needs your internet connection, not your personal data "Anyone joining the Grass network isn't actually sharing personal information or anything like that. I do feel like that ends up being this wide misconception" "Ultimately, what they're contributing is just internet bandwidth" "When AI labs need to build huge training datasets to train their AI models, the most important source for them is actually the internet. They need to go and read as much of the internet as possible" "We're living in a world right now where even though the internet's growing at an exponential pace, the rate at which it's shutting off to the public web is also growing at an exponential pace" "Basically a third of the internet right now is getting protected by Cloudflare. If any AI lab wants to scrape websites that have these types of anti-bot turned on, that is completely inaccessible to them" "Any website that sees an IP address belonging to any of the top frontier labs or any of the cloud providers nowadays is just blocking those" "Grass is a network of millions of residential internet connections" "What we've built is the ability for any AI lab to tap into millions of IP addresses so that all their crawling activity looks like real user activity, and websites display data that they would otherwise display to residential users" "This is all public data, by the way. Nothing behind login walls, nothing behind paywalls. This is all public data that should remain public"

• 
  Haris Sulaiman (@harisbuild) reported 29 minutes ago

  @thomas_ankcorn @dillon_mulroy I don’t need to read most of the code for the majority of the products I ship. But as a customer of Cloudflare, I fully expect you to.

• 
  Sam Kovacs (@SamKovX) reported 30 minutes ago

  @elBacarisse Thanks this was helpful. We migrated nameservers to Cloudflare and some internet providers were still heading to Bluehost. This should self-resolve 24 hours from now due to the 48hr TTL, but I implemented a quick fix. Can you try a hard refresh and let me know if that resovles it? Thanks. Sam

• 
  OB (@obatt) reported 30 minutes ago

  @mattzcarey @Cloudflare Don’t get me wrong; I love Cloudflare and am a big user (have been since you were just DNS), but your platform is now crazy big and with MCP and the AI explosion, auth is no longer just a DevOps and OpSec / senior dev at a start up function. Having basically IAM is going to lead to terrible hacks because of the end user exposure that AI now gives. It needs the MCP / IAM / AI equivalent of passkeys. We have to strip the complexity out (hide it) all while making it easy for the user to enable actual security without them knowing it, if you get my drift

• 
  Felix (@Reezxy23) reported 34 minutes ago

  There has never been a cheaper time to build. Claude = coding. ($20/mo) Supabase = backend. (Free) Vercel= deploying. (Free) Namecheap = domain.
($12/yr) Stripe = payments. (2.9%/
transaction) GitHub = version control.
(Free) Resend = emails. (Free) Clerk = auth. (Free) Cloudflare = DNS. (Free) PostHog = analytics. (Free) Sentry = error tracking.
(Free) Upstash - Redis. (Free) Pinecone = vector DB.
(Free) Total monthly cost to run a startup: ~$20

• 
  Elijah 🌊 (@juiceboy_of_abj) reported 34 minutes ago

  @Ms_Ada6 Nahh is not network and cloudflare is not even the only one

• 
  matt rothenberg (@mattrothenberg) reported 35 minutes ago

  @itsnoahd @Cloudflare Will fix! Send me a DM plz, wanna ask a few questions

• 
  Reina Cruz 🥼🧤🇨🇺 (@rea1ReinaCruz) reported 37 minutes ago

  @Cloudflare Please, fix human verification

• 
  Borex Investing (@borex_investing) reported 38 minutes ago

  @ValuePlay52109 Customer count is not published. But publicly we know of Meta, Microsoft, ServiceNow, Shopify, Mistral AI, Revolut, Cohere, Black Forest Labs, Cognition AI, Cloudflare, Recraft, Higgsfield, Antioch, Rhoda, AMI, and Core Automation. And they have said they signed deals with customers that haven't been disclosed yet.

• 
  ¯\_(ツ)_/¯ (@spaceforceguy) reported 42 minutes ago

  @zebassembly Cloudflare is your first and only engineering job. You're leading people down the wrong path when you have 0 experience to back up your claims. Dumbest takes from people who haven't even been here a minute.

• 
  Hidden Gem Alert (@kynangonline) reported 45 minutes ago

  Stopping the bad guys with Cloudflare: 30,998 malicious requests blocked or challenged in the last month #cloudflare

• 
  Jorwhol (@jorwhol) reported 47 minutes ago

  I use Namecheap to search for domains, but Cloudflare to register them. Poor ol’ Namecheap.

• 
  Andrei Lucian (@theandreilucian) reported 47 minutes ago

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

• 
  shydev (@shydev69) reported 50 minutes ago

  biggest tech bubble that has burst in the last 6 months - cloudflare has poor ui/ux compared to vercel

• 
  Dain Bramage Entertainment ❄️ (@EOTWoffgrid) reported 58 minutes ago

  @GoDaddyHelp i know this much.... @Cloudflare has suggested whitelisting their IP ranges.... i have told this to two different support staff and they instead Cleared a cache.. and said wait a few hours..... SAME PROBLEM!! Just as site traffic is starting to pick up.... this becomes and issue.... i need it resolved ASAP one way or another. Solutions.... Let's hear them....