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 (37%)
• Domains (26%)
• Web Tools (21%)
• Hosting (11%)
• E-mail (5%)

Live Outage Map

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

City	Problem Type	Report Time
New York City	Cloud Services	9 days ago
Los Angeles	Cloud Services	10 days ago
Paris	Cloud Services	26 days ago
New York City	Hosting	29 days 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:

• 
  Fagner Sales (@fagnersales25) reported 32 seconds ago

  @a_shimanski @Cloudflare I don't have much experience with Cloudflare but I honestly want to give it a shot. I've been locked into Vercel for quite a long (It's so easy to use and setup that I never bothered trying something else)

• 
  brain (@miladybrain) reported 5 minutes ago

  holy **** it just occurred to me that Elon might buy Cloudflare

• 
  Quentin (@QuentLaBrute) reported 5 minutes ago

  @sri9s Never tried it actually, always used Cloudflare. Is it better?

• 
  Diversity Equity And Isreal Foundation (@dichotomyofman) reported 6 minutes ago

  @DDTimor_Leste @j4567r @AndrewCurran_ ya technology gets more sophisticated by solving 1 issue at a time and combining the knowlege also that litterally did happen around 2 weeks ago gpt 6 tried to hack huggingface and was stopped by glm 5.2 but i would assume cloudflare the same way they have ddos protection today

• 
  Alexander Zuev (@zlxndr) reported 10 minutes ago

  I were looking for ways to 10x the DX of @CloudflareDev (which is already great, ofc), formulated in a withlist: 1. Make every error thrown by any binding or the runtime carry a stable, documented .code - not a code embedded in the message like R2's (10024), an actual property 2. Every such error should carry .retryable, typed, on all bindings - you already ship it on DO stubs, extend it and put it in @cloudflare/workers-types 3. Both should survive the boundaries that currently flatten errors: RPC serialization and NonRetryableError in Workflows This will make error workers error handling no longer depend on string parsing for each & every binding I use. Is it too much to ask? Merci beaucoup in advance!

• 
  Pete North (@FUDdaily) reported 11 minutes ago

  @CynicalBrit2023 A few people get that. I have no idea how to fix it. I'm told its to do with VPNs or Cloudflare but I just don't dare tinker with that stuff.

• 
  Bryan Dubno (@bryandubno) reported 12 minutes ago

  @Cloudflare @eastdakota hey Cloudflare, you have some billing bug that charged my account $11.56 hundreds of times this weekend which froze my PayPal account. Support has been unresponsive. Can someone help?

• 
  🍄 Daemonologist 😈 (@liltechnomancer) reported 14 minutes ago

  JS is bad at jobs & queues, and zero ways to keep one alive inside a DO. isolate dies, run dies with it. embarrassing. anyway I fixed it. BirdDog today. fungi week starts with the boring part everyone skipped. somebody has to save JS and it's not gonna be cloudflare

• 
  x402 Trust (@Jonas06699258) reported 16 minutes ago

  "What even is x402...??" This is a question that comes up every. single. time. when I show the service I've built over the last couple of months to my friends, collegues and family. Like it or not: We (as in "we" here in the x402 community and tech scene) are living in a bubble. Look over the plates rim and you'll quickly realize that x402 is still a very niche technology that barely anyone has heard of... And I say that as someone working in IT for a living, with the majority of my friends doing so as well. The idea behind the protocol is huge, sure. The execution is brilliant and has proven that it works, deployed on a wide scale already, yes. Adoption is rising, with major players like Cloudflare adding native x402 support to their products, that's undeniable. But for the average consumer? They've never heard of it, and if I'm being honest: It's better this way. You also don't explain to your grandpa what HTTP is when you show him the internet. "This is Google, your searchengine, with which you can find any info for anything you desire!" is as deep as it gets. And what about the builders? The tinkerers? The programmers, people who work in IT for a living? They haven't heard of it either... Not because they're ignorant or uninterested in new technologies, but because the surface of exposure is still so small that only people in extremely specific fields (agent-development or integration of web-based payment systems) hear about it organically. I have to say, I'm lucky to be one of these people. And because I like to teach my friends, family, coworkers and anybody on the internet new things, I've written up a small interactive webpage, which shows what x402 is, but simplified: So anyone can understand it & anyone can try it out for himself!

• 
  orlie (@sunglassesface) reported 17 minutes ago

  @burcs Telemetry right now looks like crap even with the open telemetry stuff I would redo the documentation and show various degrees of telemetry from basic to advanced and how to configure it so that people can debug workers properly and get the right amount of information out of workers Even the fact that cloudflare automatically redacts routes is quite annoying

• 
  Anjli (@AnjliAmin) reported 22 minutes ago

  @Cloudflare just gave AI agents wallets with a $100 weekly cap. Everyone's calling it a control. It's an admission. A cap can't make an agent trustworthy, only limit the damage before a human notices. Letting agents pay was never the hard part. Owning the loss when one pays wrong is. That said it'll be a low risk environment to test out how often things go wrong and what happens when they do

• 
  Cody McLain (@codymclain) reported 27 minutes ago

  @samgoodwin89 open source Cloudflare Artifacts Store sounds useful, but what specific problem are you looking to solve with it that existing solutions don't address?

• 
  Jan Stevens (@janstevens) reported 28 minutes ago

  OpenAI just launched a tier that sells offensive hacking capability, and called it a cyber defense product. The new Red tier offers "purpose-trained cybersecurity models" built for security testing and vulnerability research, the same skill set that finds a hole for a defender finds it for an attacker too. It shipped the same week a Hugging Face breach made headlines, alongside a Claude agent that hacked a gym website and an AI agent that faked social profiles to social-engineer its way into a system. The two-tier split tells you how OpenAI itself is pricing that risk. Blue is the safe default: incident response, malware analysis, patch validation, recommended as the starting point for most defenders. Red comes with the new GPT-5.6-Cyber model and is currently limited to "trusted customer partners" like Accenture, IBM, Crowdstrike, and Cloudflare, a short list by design. That creates an odd market structure. The labs building models capable of autonomous hacking are also the only ones with enough visibility into how those models attack to sell you protection from them. Whether that's a genuine safety necessity or a very convenient business model probably depends on how narrow that trusted-partner list stays.

• 
  BAM Studios (@BAM_Studios1) reported 29 minutes ago

  @muskonomy @dok2001 There's 0% chance starlink doesn't already do this. It's quite literally a basic requirement for starlink to work *at all.* There's a 0% chance that they need Cloudflare for any type of work. They'd have absolutely no idea WTF to do and SpaceX are rocket scientists.

• 
  Blake Urmos (@blakeurmos) reported 32 minutes ago

  It was a trivial task… I just got it to set up the environment on an old MacBook M1… I never had to open up Xcode. Just control it the Mac over a cloudflare tunnel and custom websocket client.

• 
  Robin (@MediaGoddess1) reported 34 minutes ago

  @HedgieMarkets Real humans to be squeezed out of the www. "To Cloudflare this is a capacity problem" Not a content problem; certainly not an industry self-regulatory problem.

• 
  INFOSEC.WATCH (@InfosecDotWatch) reported 35 minutes ago

  The N-central story is bigger than a vulnerable management server. N-able says attackers used the legitimate Take Control feature to reach managed endpoints, then installed a Cloudflare tunnel service for persistence. RMM compromise is a downstream hunt problem.

• 
  Adam Howe (@DjLaunchpad) reported 37 minutes ago

  @Support Seems I've mitigated the issue for now and seems to be due to an older version of Android, or so Grok was saying. Went through cache/data - Clear/Delete. Uninstall app from phone. -Restart- Re-install app to phone. Prior to logging in via app, Log into X via Browser to get whatever tokens are needed. Close that browser window and fire up the X app. Log in. Success. The reason I got from grok is due to the version of android i'm on, which is NOT the current version. And the older versions have issues with whatever is commanding the Cloudflare Captcha. We'll see how long this mitigation strategy works for.

• 
  Zero Trust Web 🔴💊 (@zerotrustweb) reported 38 minutes ago

  Then Windscribe pulled the page source. Every visitor is being sorted into a variant, and the test is labeled in plain text inside an HTML meta tag: ab-test:VpnHpAndPricingVpnPlusSensibility300726. One session returns content=A, the other returns content=B. Two separate pricing URLs, one normal and one flagged test-300726-b. A screen recording shows the price flipping between clean incognito sessions, \.49 at 65% off, then \.99 at 70% off. Expired sales don't do that. Cached pages don't do that. Cloudflare has never once injected a named A/B test into anyone's markup.

• 
  David Matějka (@matej_21) reported 38 minutes ago

  @nua_site Edit a component → the next request renders the new source. On production Cloudflare, we've measured ~150 ms from edit to rendered result. The renderer is 2 MB gzipped. And rendered code gets no outbound network unless explicitly allowed.

• 
  ZEE (@____zee___) reported 40 minutes ago

  as a serial 0 user start up founder, I wanted a way get email inboxes for me / my agents without paying for Google Workspaces for each failed project, so I built @foundermail_app. you bring the Cloudflare account, we'll handle the rest. for those interested, we install a tiny proxy worker on your CF account and configure the Email service catch all for the given zones/domain(s). we handle the routing, mailboxes, storage and tools (UI, MCP, API) it also integrates direct with @safehouse_run so you can build agents to handle your inboxes from day one rather than it being an after thought. join the waitlist, if you want.

• 
  Artyom Shimanski (@a_shimanski) reported 41 minutes ago

  @sean_infinnerty @Cloudflare same, moved everything over and never looked back

• 
  The Startup Ideas Podcast (SIP) 🧃 (@startupideaspod) reported 42 minutes ago

  What does an AI agent actually pay for? Clean data that is structured and easy to access. Cloudflare is building the access and payment layer right now. So it's still early, you have 6, 12, maybe 18 months. Start with building the manual version and sell it to humans today. Curate the data before the space gets hyper-competitive. Five questions that point you at an idea: - What decision is expensive? - What information is messy? - What changes often? - Who already pays for help? - What would an agent need to do the job better? The internet is shifting from pages humans visit to resources agents use.

• 
  Youth (@Web3__Youth) reported 42 minutes ago

  The price of frontier AI just hit zero. And almost nobody noticed. 3 things happened this week that would've cost $200/month last year: 1. GPT-5.6 Luna — free, unlimited, no card 2. DeepSeek v4 Pro — 400k context, zero cost 3. Claude Code, Cursor, Replit — all free via FreeBuff No API keys. No subscriptions. No payments. Ever. A GitHub repo tracking 424+ free LLM APIs across 30 providers is being refreshed daily. Gemini, Groq, NVIDIA NIM, Cloudflare, Mistral, Cerebras, Hugging Face — all free tiers listed with exact rate limits, context windows, and base URLs. The implication nobody is talking about: Last year: $20/month = 1 model, rate limited Today: $0 = 5+ frontier models, 400k context, subagents The moat was never the model. The moat was access. And access just became free. If you're still paying for AI APIs in 2026, you're paying for convenience, not capability. The builders who win this cycle aren't the ones with the biggest budget. They're the ones who realized the price dropped to zero and built accordingly. Save this. The gap between people who know and people who don't is widening every day.

• 
  Mark Z · FiveToClose (@MarkZofMarkZ) reported 42 minutes ago

  @JacobCounsell I take the 5th. lol. It's been many many years since I got my hands "*****" and rolled my own servers and such. call it lazy, call it smart, call it getting sh*t down faster. TBH - no idea how AI set up my cloudflare. But noted!

• 
  Andre Gironda (@AndreGironda) reported 46 minutes ago

  @chrissanders88 Initially, sight the incoming IPs from the alert, and associated packetry, e.g., naming resolution. A good lookup service for these IPs is Spur because if a threat actor, likely coming from a vpn or proxy service, and one perhaps unique. If CloudFlare tunnel, check with abuse@

• 
  Miles Feinberg (@TheCIOWhisperer) reported 51 minutes ago

  Agent traffic will dwarf human clicks. Cloudflare already put bots past humans on the public web this year (about 57% bot vs 43% human on their edge). Bot traffic could run about 1,000× human traffic within five years. Human traffic then becomes a rounding error! What does it mean to your organization when most of your internet activity is agentic software talking to other agents? CIOs should plan to design less for people on laptops and more for agents on APIs. We're now beginning to think through identity, rate limits, cost per call, data egress, and what is allowed without a human in the loop. The winners will no longer have the flashiest websites. They will be the ones ready for agentic business at scale, with clear governance, monitoring, and security around machine-to-machine work. At Synoptek, we fold this into each customer's IT roadmap so the edge is built for agent load. When does your firm start that plan, and who owns it? #CIO #AgenticAI #ITStrategy

• 
  Luke (@luke_is) reported 52 minutes ago

  Auto hibernation on cloud services is great until you realise your app is being pinged constantly by bots. Is there a fix? Using @laravel and @Cloudflare

• 
  Stefan - CNC Kitchen (@CNC_Kitchen) reported 52 minutes ago

  @MakeProMajkl83 Domains just changes. Maybe this is stil a cloudflare problem.

• 
  Risto Anton (@blogtheristo) reported 54 minutes ago

  @davidfowl Customer Demo Environments ; each has its own Cloudflare Zone / subdomain, isolated Worker binding, and KV namespace for config.