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.

• Domains (36%)
• Cloud Services (29%)
• 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	Hosting	2 days ago
Manchester	Domains	22 days ago
Angers	Cloud Services	1 month ago
London	Domains	1 month ago
Noida	Hosting	2 months ago
Jewar	E-mail	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:

• 
  Nick (@heyiamnick_) reported 10 minutes ago

  I needed better media storage for the library. Cloudinary was fine, but the credit-based pricing could get expensive as bandwidth grows. So I moved everything to Cloudflare R2: - Around $0.015/GB - Zero egress fees - Works for images, videos, ZIPs, PDFs, and backups The only problem... R2’s UI is painfully basic. I had to open files one by one just to preview them or copy the URL. So I asked Claude to build a custom media dashboard. Now I can preview everything, browse folders, copy URLs instantly, and automatically sync local media to R2 after every *** commit. Cheap storage + a custom AI-built interface. That is the kind of AI development that is actually useful.

• 
  Paul (@soypaulco) reported 10 minutes ago

  I genuinely appreciate @Cloudflare while working on an demo project using D1, I forgot to flip an ENV and somehow wrote ~142M rows in overusage. An impressively rookie mistake. Support cleared the charge after I explained it. Huge shoutout to Jacob from support for helping me!

• 
  pagal🦋 (@ssahillppatell) reported 10 minutes ago

  there's no "Create a case" button on @Cloudflare support page. ?????

• 
  Brais Calvo (@braiscv) reported 11 minutes ago

  @antonosika On Cloudflare, mainly for Zero Trust. And honestly, the switch to token based billing for the Cloud features has been pretty terrible. Another reason for hosting it externally is to have more control and transparency over the stack.

• 
  Sunder (@talk2sunder) reported 11 minutes ago

  Our Teams bot went silent. Zero errors. Zero logs. Azure said healthy, Cloudflare said healthy, our servers said healthy. Slack worked fine. Same bot, same endpoint. Gave it to Claude (Fable 5). It pointed the bot's endpoint at a plain webhook listener. Microsoft's messages showed up instantly. So Teams WAS sending — something about our edge was unreachable. One openssl probe later: our CDN required TLS 1.3, and Microsoft's Teams delivery fleet still speaks TLS 1.2. No logs because of handshake failire. Fable 5 rocks.

• 
  Naoris Protocol (@NaorisProtocol) reported 13 minutes ago

  @Weaver_Labs @Cloudflare Fair parallel. Telecom had to retrofit a live network with millions of legacy devices already in the field. A chain built with ML-DSA-87 at genesis skips that coordination problem entirely, there's no fleet of old signers to bring along.

• 
  lolnotacat (@lolnotacat) reported 13 minutes ago

  @TheOtherMassie I just got a verification. Try a couple different servers and regions. The IP you're using was probably flagged by cloudflare or other DNS service.

• 
  BruzWJ (@BruzWJ) reported 16 minutes ago

  @zoltanszogyenyi @Cloudflare auto audits are neat but the fix list feels sus if it skips the accessibility issues that actually bite users 🤔🧩👀 does it catch the real a11y stuff or mostly contrast + alt text?

• 
  Glenn 'devalias' Grant (@_devalias) reported 19 minutes ago

  @nickgraynews @computefinx The help docs also mention D1 and R2, which are Cloudflare things

• 
  KaiseiDeer (@Kaisei_Daloren) reported 21 minutes ago

  **** cloudflare I can't signin to any of my accounts and have tried all the suggested things I have to use incognito mode now.

• 
  Payam (@im_payam) reported 24 minutes ago

  This is the FASTEST way I've found to ship. Built for coding agents in terminal. Here’s the setup: - Get a cheap VPS (16 or 8GB) - Install Tailscale, block ALL traffic except Tailscale - Install Claude Code on the VPS - get a domain on Cloudflare - grab Cloudflare API token and pass to Claude - ask Claude to connect the domain to the server through Cloudflare's proxy (do not open ports) - Setup your site with Next.js, or vanilla JS + HTML - Use SQLite for database. (it's just a file) - ask claude: "Setup Restic to nightly backup DB and .env to R2" That's it. you are now the fastest and less error prone setup to build real products with agents. - Zero platform lock-in - no crazy bills from Supabase - no timeouts from Vercel - scales to millions with no limits - build it, restart service, it's live! just copy paste this to Claude to get started.

• 
  ₽ⱤΞΞ₮Ⱨ△M ₭Ɏ△И△M (@pkyanam) reported 24 minutes ago

  @southpolesteve @southpolesteve will support Workers / Drop from Cloudflare too for agent generated artifacts/sites!

• 
  JamaisVu (@bitten_2wice) reported 25 minutes ago

  @TetoLuuver @n1ght_watch3r_ @netangelyuri "Connected without internet" is so vauge lol pinging cloudflare usually narrows it down

• 
  Rob Terrin (@RobTerrin) reported 26 minutes ago

  @yrechtman Kernel access was really just useful for EDR (Crowdstrike and SentinelOne) and there have been multiple big exits since that gen like Cloudflare, Netskope, Rubrik, Wiz and Armis. If the labs lock down agent access, no cybersecurity company will become the next public company platform like Crowdstrike.

• 
  OnlyAWassie (@onlyawassie) reported 26 minutes ago

  @ztrader369 Legal Problems with robinhood copyrights the got reported to cloudflare and cloudflare disabled everything

• 
  rejex visions (@rejexdotvisions) reported 26 minutes ago

  @samirande_ sheesh that's good but bad too, cloudflare would be better ig

• 
  SecureChap (@SecureChap) reported 26 minutes ago

  Proofpoint researcher Rachel Rabin published a technique that turns Microsoft Entra ID error codes into a credential oracle. An attacker posts to the /common/oauth2/token endpoint using the ROPC flow and a random UUIDv4 client_id that matches no registered application. Entra replies with AADSTS50034 for nonexistent usernames, AADSTS50126 for valid usernames with bad passwords, and AADSTS700016 when both username and password are correct. The sign-in log records only the GUID; the application name field stays empty and no successful sign-in event is created. Campaign UNK_pyreq2323 ran from January through March 2026 using python-requests/2.32.3 from AWS. It generated 700,000+ client_ids by mutating the last six digits of the Exchange Online first-party app ID and tested over one million accounts across roughly 4,000 tenants, locking out 28 percent of targets. A second campaign, UNK_OutFlareAZ, began in December 2025 via Cloudflare and enumerated alphabetically from wordlists using 3.7 million fully random UUIDv4 client_ids against two million users. Two unrelated actors reached the same client_id spoofing method independently. The same observable behavior that powers normal authentication also supplies the validation signal.

• 
  Eric Taylor (@bcs_erictaylor) reported 32 minutes ago

  Date: 7/15/2026 Here is an update on our updated information on our proactive efforts to heavily impact the Kali365 platform. As expected, the user "Octopus King" is working hard to rebuild is infrastructure, but we have the fingerprinting locked in and will continue our efforts. At the time of this blog posting, we have submitted an IC3 report but honestly not expecting much. I believe it will be up to companies like ours to keep submitting takedown requests as fast as we see them to limit its operations. Below is the information Believed owner and creator of Kali365: Handle: Octopus King TG: @tentacle_network Completed 'Takedown Requests' submitted to Namesilo FQDN: updateteampanel[.]xyz Completed 'Takedown Requests' submitted to BL Network: IP Address: 199.21.221[.]21 Pending 'Takedown Requests' submitted to BL Network: IP Address: 162.33.178[.]105 IP Address: 72.5.43[.]195 Pending 'Sinkhole Requests' submitted to @Cloudflare FQDN: updateteampanel[.]xyz FQDN: privatetoken[.]app FQDN: servoquil[.]org FQDN: vuredonte[.]org FQDN: yalmorind[.]org FQDN: ondrevail[.]org FQDN: caldivore[.]org FQDN: mvpaffiliatecz[.]site FQDN: vuredonte[.]org #Kali365 #CTI #threatintel

• 
  Brad Aldcroft (@baldcroft) reported 33 minutes ago

  Locked out of cloudflare account due to likely email supression. Going around in circles trying to lodge a ticket for help on your site without account access. @CloudflareHelp @Cloudflare

• 
  VESKALD (@veskaldofficial) reported 38 minutes ago

  Trading keys can't reach us even by mistake. A key with trade or withdrawal rights is rejected at submission — never stored. And there's no server IP to whitelist it to: our infrastructure sits behind Cloudflare with no public address. The exchange itself won't let it work. Not a policy — an architectural dead end.

• 
  Anirudh Oppiliappan (@icyphox) reported 39 minutes ago

  @threepointone @AKuederle Not sure where all this anger is coming from. The folks on Bluesky (and me here) were poking fun at the banner text—which was apparently *meant* to be tongue in cheek! Wasn’t clear. Nobody’s trying to cancel Cloudflare. Let’s all chill out for a bit.

• 
  SoothSpider 🇨🇦🍁🧡真🔬💻Ω 🐶😼🌎 (@SoothSpider) reported 39 minutes ago

  @45Homelab Can I stick a PVE (zfs mirrored) at my out-of-State/Province friend's house configure some basic services with OOTB easy HA/redundancy? 🤔 Can I put that behind a $5/month CloudFlare load balancer? Can I spin up a new service from scratch knowing nothing in a few hours? 🤔

• 
  lekeeeeeeeeeeeeeeeeeee! (@joshuadeleke) reported 42 minutes ago

  The craziest thing you can do is trying to access cloudflare with Airtel, the network too useless

• 
  Cosmin Dolha (@CosminDolha) reported 45 minutes ago

  If Apple were to decide to build a battery-powered SBC geared toward edge AI (inference), and just put the M1-M2 chip with 8 GB Ram, it would obliterate anything on the market, even if priced a bit higher, let’s say in the range of 300-400 USD. There are no SBC that can achieve the latency of Apple Silicon for local edge AI. But you don't really have to wait for Apple to do that (chances are, they won’t), since most of the intelligence for your edge AI projects can be easily streamed from a Mac mini to any device, including MCU (ESP32, etc.) and including outside your Network by using Tailscale or Cloudflare. Also, you can buy a refurbished Mac mini with M1, at around $280– $350. You won’t have the GPIO, if you really need them, you can use a USB-to-GPIO module, but this is for inference, so you would have your gadget built with your end-client choice, an ESP32, or some low-powered Linux, and stream inference results to it. Apple has won the edge AI hardware race; its not even close, price/performance. Maybe they don't really have to win the software stack, since that will be commodities.

• 
  Emely (@EMYokay) reported 47 minutes ago

  @ATKINGDOM @NoirZarina is your site down? cant log in. error 502 cloudflare

• 
  Maverick (@Mavericks100xs) reported 50 minutes ago

  It’s over for cash-cat:native Chinese blockchain sleuths have uncovered the following: ‘NOXA Dev ***** History" After the incident erupted (especially post-downtime + new launch halt), the Chinese community quickly unearthed Amun Phantom's past record, with the core accusation being **"veteran rug pull playbook."** Main sources are posts from active Chinese KOLs/communities (e.g., @DiYi_Community, etc.), claiming "people who know him are well aware." - Key exposé points: Not his first big project: Two years ago (around 2024), he built a product even hotter than NOXA this time, then rugged at peak hype, allegedly draining that chain's liquidity pool dry (claims of "chain pool leader 3w ETH," possibly 30,000 ETH level, with community debate on exact figures). - Patterned operations: Every time a new chain heats up, he spins up a similar launch platform/product, quickly harvests traffic and fees, then "exits" or rugs. In 2025, multiple "exited products" of his popped up on new chains. - This time a "soft rug" not hard rug: Originally geared to straight-up bolt, but with too many bagholders this round—scale too massive (fees too high, user base huge)—direct rug would've blown up, so he opted for "soft rug" strategy: Website "issues" (Cloudflare IP block), halt new launches, shift to decentralized frontend, hand fees fully to creators, then slowly fade out. - Personality/Style Critique: Community calls it "deep-seated foreign scumbag traits, no vision for real growth," akin to some early infamous but controversial project vibes. These exposés are currently mostly community word-of-mouth + historical pattern inference, with no full public on-chain evidence chain or article yet (some say a detailed timeline post is coming). But since Amun Phantom is anonymous/semi-anonymous, historical project links rely mainly on community memory and behavioral pattern matching.

• 
  Sandip Dev (@devsandip) reported 51 minutes ago

  holy **** @Glassdoor and @indeed all i wanted was to read a few interview questions for a particular company on glassdoor. now i have spent the last 10 minutes, filling 2 otp, 5 cloudflare capthas, yes/ok/continue on like 2 dozen screens and you still want more info. **** you

• 
  Chibugo | AI automation (@__chibugo) reported 55 minutes ago

  there's a trend on instagram right now: everyone's building their own version of Tony Stark's Jarvis; a voice agent that runs like an actual OS. what nobody shows is everything people overlook when building one. so here's that part, since I built it myself. 🎯 the plan on paper vs. the plan that shipped the amount of planning you need to do is insaneee. first version had each agent loading one tool, fetching, verifying, then unloading before the next; all built around not blowing past a token budget. that's the part people underestimate: you don't know your constraints until it's running. once it was live, token budget stopped mattering. speed and cost-per-call did. so the whole approach changed: instead of mcp's, every agent just calls its tool's API directly, does the work in plain code, and only asks the AI to write the sentence at the end. 🎯 scope one AI that "does everything" sounds impressive but could be a nightmare to debug. so i split it into 5 sub-agents, each only knowing its own lane. a router decides which lane(s) a question touches. these sub-agents then report to the main agent orchestrator. similar to a team lead and team members 🎯 prompt chaining integrated ElevenLabs for voice, and a single voice reply isn't one AI call; it's a handoff, several times over: hear the words → figure out what's being asked → pull the data → write the sentence → speak it. every handoff adds seconds and cost, which can lead to latency. one reply once took 31 seconds. pulled the logs instead of guessing: a wasted double-check on an expired token, a slow handoff to an outside service, plus the normal chain. fix is running sub-agents in parallel and timing each call. 🎯 tokens and prompt caching each agent's instructions get cached, so it's not re-reading the same manual on every call. what never gets cached is live data; for instance, caching a bank balance is just caching a wrong number the moment it changes. that same cost-awareness came back around differently later: the AI account ran out of credit for a few hours, and every request failed with the same error. 🎯 local vs. web everything gets tested on a local copy first; headless browser opening the dashboard, checking for errors, a test message round-tripping through voice before go-live. if you're handing this to someone else, though, you commit, push to GitHub, and host it. for mine, i used Cloud Run for the brain, Cloudflare Pages for the screen. 🎯 vault write-back most builds only go one direction: ask, answer, forget. this one writes back because every full briefing gets saved as a dated file into a notes vault so the agents can keep training themselves with the data. if you ask a follow-up an hour later and it still knows, because that memory is shared across every way you talk to it. used Obsidian, synced through Drive. 🎯 security once, a message could've gone somewhere it shouldn't have. that only needs to happen once to be a problem. so now there are three checks before anything goes out: - it can never post to certain places, - it has to prove who it's speaking as before it speaks for someone, - and tests confirm both of those actually work. separate from that, I went through every access key this thing has and asked, "does it really need this much access?" a few did not. 🎯 the checks after every update, it runs a test and asks does a voice message go through? does it understand a normal sentence? does the screen load with no errors? also implemented something we call "error logging" in automation, but i call this "diagnostics" or "agent health status," which checks in every 15 minutes on its own, making sure the data is still updating. if it's not, it sends a warning without anyone needing to notice first.

• 
  Random Crypto Chad (@RandomCryptoCh1) reported 56 minutes ago

  @AlbertMacGloan @RobinhoodCrypto @Noxa_Fi u tripping? noxa is larping that its cloudflare issue, when clearly they dont have CF ns even set up, everything on RH works, except them.

• 
  Trevor Hudson (@trevhud) reported 59 minutes ago

  With cloudflare tenant separation (one D1 per customer for example) you could just start giving read access to your customers which would go a long way for trust and integrations.