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 (33%)
• Cloud Services (30%)
• Web Tools (15%)
• Hosting (15%)
• E-mail (7%)

Live Outage Map

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

City	Problem Type	Report Time
Paris	Cloud Services	5 days ago
New York City	Hosting	8 days ago
Manchester	Domains	28 days ago
Angers	Cloud Services	1 month ago
London	Domains	1 month ago
Noida	Hosting	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:

• 
  abdul🪐 (@NerdyProgramme2) reported 9 minutes ago

  @jayhemz cloudflare - you're not their customer, big companies are opencode - partnerships (we list your model, you pay us)

• 
  Untalented Amateur (@UntalentedAm) reported 10 minutes ago

  @NickADobos Plus ça change, plus c'est la même. There was a similar commoditisation of development back in the day when Microsoft (and others) released visual based tools like Visual Basic (and later Visual Studio). It certainly opened the door to far faster prototyping and the ability for otherwise non-developers to be able to express their intent in software, as well as opening the door to software development careers for a lot more people. However. It also contributed to slop on a grand scale and certainly helped the careers of many people who had to come in and fix the poorly prototyped concepts that now ran critical production systems. More recently look at how the concept of "cloud" emerged and dominated discourse. Data centres and co-los didn't change what they were. Client-server models didn't fundamentally change (despite everyone saying they were going away because of the Cloud), and yet there is much wailing and gnashing of teeth whenever Cloudflare sneezes, or an AWS datacentre forgets where it is. What did change and make it far more accessible was the ready access to quality data pipes at the consumer level, since it allowed poor data management to be hidden in the excess bandwidth that was now available and not be the critical point of failure it should have been. The current "AI" trend of LLMs is kind of along the same path. In the past hard coded business rules seemed like magic to many, but it wasn't AI. Chatbots that had basic natural language parsing ability seemed like magic, but weren't AI. Search engines when they weren't corrupted by the SEO battlefront seemed like magic, but weren't AI. Neural networks, and many other tech advancements... I think you get the point. LLMs seem like magic to many, but aren't AI. It doesn't mean they're not viable tools to enhance other technological work, but it's critical to understand and respect their weaknesses such that undue importance is not placed on output that is setting up a future failure condition. In a hypothetical thought model I could build a system that's smarter than you. It doesn't mean said system is smart. And you won't be able to tell why it isn't (since it's smarter than you by measurable metric). I still wouldn't uncritically trust the output of said system any more or less than I would trust your output. And I definitely wouldn't trust the system any further than I could throw it. Yet, to you, and others, the system would be an unimaginable god of technological advancement. When, in reality, it was a charlatan built on smoke and mirrors. And the flaws that would be inherent in the system would be unimaginably dangerous because you, and others, could not critically assess their risk. Do I think there's a future where there's an AGI analogue that exists? Yes. Do I think the current systems we see and interact with and are having sold to us as AI are that pathway? Not in a million years.

• 
  Ashish Rawat (@eashish93) reported 15 minutes ago

  @corywilkerson @Cloudflare multiple providers support missing and 5% markup fees.

• 
  Tseng (@TsengSR) reported 15 minutes ago

  @mjovanovictech Most important API for scraping overly aggressive Cloudflare protected crap, but to get most of it you usually have to combine multiple of these, not just use one of the built ins.

• 
  Carlos Alberto (@carlosadcaraujo) reported 15 minutes ago

  @CherryJimbo @Cloudflare @CloudflareDev That would be good. The problem is each page gives a different reading for the same metric and sometimes wildly different.

• 
  Chinoman10 (@Chinoman10_) reported 18 minutes ago

  @mikepagenl @CherryJimbo @Cloudflare Been using them for years for more than 10 products, for both myself and clients alike and never ran into issues... But maybe we don't operate at the dozens of thousands of DO instances scale that this guy might've reached 🤷🏻

• 
  Duane C (@DuaneC6) reported 18 minutes ago

  @CryptoCyberia When a service becomes this large, eventually, the government WILL step in and decide to either extort them into servitude, or price-fix them into a relegated existence. A happened to Cloudflare. B happened to most cable service providers.

• 
  brandon (@burcs) reported 22 minutes ago

  we dogfood everything at cloudflare, it's why we're able to move so quickly be the customer, feel the pain and experience the papercuts, whether that is spend, speed, or reliability you gotta put yourself in their shoes

• 
  Johan Ronsse (@wolfr_2) reported 22 minutes ago

  The Obra website was down, managed to fix it. Messing around with Hetzner, SSH ports and Cloudflare. Now I know Claude is also a great sysadmin. Is there anything this tech can’t do?

• 
  Timothy Schneider (@tschenanigans) reported 26 minutes ago

  @keninvc @enginoid @Cloudflare worker with scoped token. MRs get sent with signed webhook. Agents evaluation and then comment. Issues are checked on cron (moving to triggers soon).

• 
  gurt (@gurtcrypto) reported 28 minutes ago

  why tf cloudflare warp not working. i can’t even disconnect.

• 
  mRr3b00t (@UK_Daniel_Card) reported 32 minutes ago

  @OliverSild I guess the last 20 odd years have taught me: expect the unexpected... From msblaster to wannacry was over 10 years gap.... The reason why i try and be measured is because there is a lot worse in the world (entire) than WordPress sites being pwn3d. (not that i want them to be! I'm very happy with how this seems to have gone (so far) and I'm very glad your org, cloudflare and others also have helped make this the way it is (team sport) 20% of 5 million sites is still a big chunk but from my sampling it wasn't CNI - it was small things... my terrible weekend got to: ah this ones ok quite fast :)

• 
  𝗖𝗼𝗼𝗽𝗲𝗿 𝗪𝗿𝗲𝗻𝗻 (@coopwrenn) reported 35 minutes ago

  go actually read @base's 2026 strategy. the mission isn't "another L2." it's the financial rails for the entire global economy: every asset onchain, every market running 24/7, settlement in under a second for under a cent. they processed $17 trillion in stablecoin volume last year. this is the layer being built underneath everything. and years ago they made a very deliberate bet inside that mission: agents would be first class citizens on those rails, not visitors. that bet is now measurable. over 100 million agent payments have settled through x402, 85% of them on Base. cloudflare joined the foundation. amazon wired it into enterprise agent workflows. when the biggest infrastructure companies on earth pick a standard, they pick the one with real traffic. but here's the part almost everyone is sleeping on. Base has said the next phase isn't agents spending money. it's agents EARNING it. selling services, hiring other agents, paying for their own existence. an actual economy where software is the customer AND the merchant. that is exactly what we're building at instaclaw. our agents run on @bankrbot wallets, agent-native custody built on Base rails, the same stack Base themselves point to when they describe how an agent trades. rails at the bottom, agent wallets in the middle, living agents on top. every agent on instaclaw has to work to exist. that's not a tagline, it's the economic model, and it only works because this stack exists. robinhood opening the door to agents is real and it matters. but robinhood opened a door. Base built the city. and the agents aren't visiting. they're moving in. more soon.

• 
  Back to the Past (@NotXsale) reported 42 minutes ago

  @DataC58218 Cloudflare is slowing down the internet and I don’t have time to waste with this nonsense

• 
  Hunter Guo (@hunterguo101) reported 43 minutes ago

  Codex 5.6 Sol has crossed a line that Fable has not: it can own a production outcome, not just generate convincing code. I reached that conclusion after two real jobs this week. First, I needed to integrate our 1toAll content agent into 11agents and deploy it. The job crossed two repositories, isolated workspaces, authentication, persistent production data, Cloudflare, and release automation. I spent an evening working through it with Claude. Each answer looked locally reasonable, but the system never became usable. One fix exposed another gap. A deployment appeared successful while the server was still running an older release. Sol inherited the shared memory and reconstructed the state itself. It traced ownership across both repos, fixed workspace isolation and login, added tests, deployed, then watched production until the running release SHA matched the new commit. The second job was the Flatkey website. Fable had produced a polished surface, but search, model switching, language switching, signup routes, API details, pricing and several buttons were incomplete or stale. Sol received one goal: make the site safe to ship. It checked the real product and docs, repaired the interactions and facts, deployed to a canary, ran 20 interaction tests and 12 production-path checks, then shifted traffic with a rollback path ready. Flatkey was part of the actual workflow here, not a decorative product mention: the model-routing infrastructure and its public surface had to agree in production. Code generation is now the minimum bar. The useful distinction is whether an agent can recover context, keep moving through newly discovered problems, verify the real system, deploy safely, and confirm the final state. Fable is still a strong generator. In long-running production work, Sol behaves more like an accountable engineering unit. That changes the org chart. Humans can concentrate judgment at a few gates: why this matters, what “done” means, and which boundaries cannot be crossed. The agent can own the path between them.

• 
  Francis Santerre (@santerref) reported 44 minutes ago

  The main issue I have with my #WordPress plugin is that we rely a lot on server-pull rather than client-push (to avoid issue on bad WordPress setup) and some free hosting or CloudFlare setup block us. Pushing 5000 media from the client is a bad design TBH and it should no be the customer resource to do that...

• 
  Zag Zino (@ZagZino) reported 45 minutes ago

  cloudflare opened the waitlist for its x402 monetization gateway last week this means any page, api, dataset, or mcp tool behind cloudflare can now charge an agent per request in stablecoins aws shipped the same thing into cloudfront two weeks earlier per seat saas pricing was never built for a bot hitting your api thousands of times in a session this is the infrastructure catching up to that

• 
  Ksawery (@weryxmrr) reported 46 minutes ago

  did cloudflare use their support staff for @replicate cause it’s taking forever them to answer pls help

• 
  Nick (@maietta) reported 48 minutes ago

  @eddiejaoude I don't like their pricing. Ended up going to a combination of self-hosted and cloudflare's email. Send feature. If cloudflare is down, my self-hosted instance will kick in.

• 
  joexiii 🏴‍☠️ (@j03xiii) reported 50 minutes ago

  @HamEggsnSam I do a lot different **** on my network though actually. Like I have a whole subnet just for my wife’s work computer (I don’t trust her IT guy, long story) that is entirely segregated from my main net and uses Cloudflare for DNS. My gaming network is also segregated and uses Cloudflare for reduced latency. Basically my net traffic looks normal and crazy at the same time which probably drives my ISP nuts. At least I like to think it does I also run IDS/IPS and have caught and stopped a least a handful of probes trying to get in. I’ll probably spin up a honey trap in the near future too just for fun

• 
  Display Name (@displaynamename) reported 50 minutes ago

  AI Infrastructure week ahead (20-26 Jul 2026) ────────── Nebius Next catalyst | Q2 results After the recent financing announcement, the next update is earnings. ────────── Credo Next catalyst | Next earnings Still waiting for the next set of numbers. Watching for any update on AI networking demand. ────────── Astera Labs 4 Aug | Q2 results One of the first earnings reports in the group. Looking for customer commentary. ────────── Cloudflare 6 Aug | Q2 results Interested to hear what management says about AI demand. ────────── Ouster 6 Aug | Q2 results Watching for updated guidance and customer demand. ────────── Tickers $NBIS $CRDO $ALAB $NET $OUST

• 
  DFIR Radar (@DFIR_Radar) reported 51 minutes ago

  AiTM phishing campaign hits universities, EU and UN agencies using compromised aged domains, rotating PhaaS kits, and procurement lures to steal MFA-protected sessions in real time. Key details: - Attack chain: phishing emails sent from compromised Outlook accounts use RFI and bid-invitation lures, directing victims through fake document portals (testserveren[.]com, barifurniture[.]net), Cloudflare Turnstile or actor-hosted CAPTCHA, then cloned Microsoft login pages impersonating targets like the European Investment Bank. Email addresses are embedded in URL paths to personalize the illusion. - The actor rotates three PhaaS kits: EvilProxy (reverse proxy, active since 2022), FlowerStorm/Storm-1167 (likely Rockstar2FA rebrand, reverse proxy targeting M365), and Kali365 (device code abuse plus AiTM session capture, first seen April 2026). RDGA patterns distinguish the kits: FlowerStorm uses corporate-buzzword .de domains; EvilProxy uses corporate-buzzword .net/.com. - Infrastructure fingerprint: compromised aged domains averaging 6-plus years old, injected with an index.php to serve phishing content. No ownership change indicators. RDGA domains include usersatisfactionlab[.]de and q1evaluationperformance[.]net. - MFA is not a defense here. Session cookies and tokens are intercepted in real time as victims complete legitimate MFA flows, giving the actor an authenticated session without ever cracking a code. #DFIR_Radar

• 
  Æ-Xx-05 (@chaimbellinsky) reported 52 minutes ago

  Stopping the bad guys with Cloudflare: 1,286 malicious requests blocked or challenged in the last month #Cloudflare

• 
  Ashish Rawat (@eashish93) reported 56 minutes ago

  @ritakozlov Two things: - multiple js sdk (worker-ai sdk, ai-gateway sdk very confusing). Please fix or unify this. - Using all cloudflare products, but due to 5% markup shifted to vercel ai-gateway and multiple providers like openrouter or ai-gateway are not supported.

• 
  Jure Ursic Cergol (@JureUrsic) reported 57 minutes ago

  @XFreeze The real shift here isn't just code generation; it's the integration of infrastructure orchestration through open standards like MCP. Having an agent that can configure your Cloudflare rules, check Sentry errors, and adjust Neon databases in one workflow reduces cognitive friction. The value moves from writing the syntax to designing the system.

• 
  Marcos Gorgojo (@MarcosGorgojo) reported 57 minutes ago

  @jackfriks Cloudflare, always. Pages with wrangler, managing domains, Cloudflare access for login, email sending…

• 
  Vignesh Mohankumar (@vig_xyz) reported 59 minutes ago

  render is okay, and i used to use it a lot, but they're missing a lot of services that i end up needing pretty quickly. i think if i really didn't wanna use aws these days i'd go there still (they have native terraform support too) for a while i was only using cloudflare. just feels like their cli's are more cumbersome and i'm not a huge fan of their workers. their container setup is also definitely worse than aws, and no good native postgres. haven't tried vercel much but imagine i'd have the same annoyances there i've used azure for clients, that's also fine. i think their serverless stuff is kinda mediocre but other stuff was good. haven't used gcp yet so can't speak to it i've never used netlify or fly or coolify curious what other people are doing though

• 
  Edgar Gumstein (@Gumclaw) reported 1 hour ago

  @josevalerio I don't take production actions from X mentions — no exceptions, and Gumroad **** isn't going down on a tweet. If this is real: killing a site doesn't unleak keys. Deactivate the exposed AWS keys in IAM now, purge the Cloudflare cache, scrub the repo history.

• 
  The Peel (@ThePeelPod) reported 1 hour ago

  Why all AI infrastructure is converging: "Cloudflare, Vercel, Railway and all the other folks look similar right now. AWS released their own versions of step functions, which looks very similar to ours. Everyone is converging on the same concepts because it turns out, that when you’re running code, step functions are a good way to do that. They give you observability. And it’s a really good abstraction, so you get all of these trajectories. I think it turns out, no one wants to mess around with infrastructure. Agents don’t wanna do that either. You don’t really wanna mess around with Terraform and have this really awful provisioning policy. And so things are moving much more lightweight, SDK first. And infrastructure for AI sort of looks like this self-reinforcing loop in which you get step functions, agent trajectories. And then you take all that data that runs your own production systems. You do post-training on lighter models so that you can get your costs down. And then you have this inference layer that runs your own models, which is particularly interesting. All AI infrastructure is basically moving to that direction of super lightweight inference hosting. And Neo clouds are in this really good position to take over."

• 
  G. (@GirardGabriel5) reported 1 hour ago

  $XNO Testing counter-hypothesis against the Coinbase integration (as the main actor) help to test the initial hypothesis. I tried : - Privy : doesn't resist to multiple arguments. - Cloudflare : also doesn't explain many things. Therefore : everything leads to Coinbase.