Every other week is CrowdStrike now: 2026's self-inflicted AI outage subscription
Select a badge for its explanation and points.
All badges 46
Company conduct 11
Money and management 14
PR and hype 14
Monetisation, platforms and ownership 7
Remember when CrowdStrike felt like the end of the world? Back in July 2024, one botched update blue-screening half the planet felt like a freak, once-in-a-decade disaster. In 2026, it feels like a quaint nostalgia trip! Practically every other week this year, another massive chunk of the internet has keeled over in broad daylight, making 2026 comfortably the most outage-ridden dumpster fire we have ever seen.
Just look at the scoreboard. Control Plane's outage tracker had already logged 19 major cloud meltdowns across AWS, Microsoft Azure, Google Cloud and Cloudflare by mid-November, smashing past all 16 recorded in 2025 with weeks left on the clock. Cisco ThousandEyes logged 419 global network outages in a single week this month (down from 450 the week before!), compared with just 128 in the exact same week last year. And in one glorious six-day pile-up this month alone, Microsoft 365, the FAA, Workday, AWS, JPMorgan Chase, United Airlines, Apple and even the NSA's classified internal network all tripped over themselves and went dark. (Control Plane, PBS News)
Could it be!? Has the mythical AI machine god finally woken up to pull the plug on humanity!? US Homeland Security Secretary Markwayne Mullin certainly got caught up in the sci-fi mood, going on television to float an unnamed tech buddy's theory that "self-fixing" superintelligence was overrunning FAA systems (a claim the FAA and Transportation Secretary promptly shot down). Sadly, if you were hoping for a grand robot uprising, you are stuck with something infinitely dumber. AI is right at the centre of this mess, alright, except there is no malevolent god pulling the strings. It is just the most comical display of self-inflicted corporate stupidity you could ever expect from an out-of-touch Silicon Valley bubble drunk on its own pitch decks. (WindowsForum, Newsweek)
Microsoft's monthly outage subscription:
You genuinely could not ask for a funnier poster child than Microsoft. One of the biggest companies on earth, charging half the planet through the nose for cloud subscriptions, has managed to suffer a public faceplant practically every single month of 2026:
January & February: A nine-hour Microsoft 365 outage during routine maintenance, followed by a multi-region Azure control-plane stumble and a 20-hour West US blackout.
March to June: Smaller Exchange Online wobbles peppered across spring, an almost 12-hour Azure East US regression in April, a 22-hour Azure West US 2 outage in May, and two separate June outages where Copilot itself fell flat on its face.
July & August: A nearly five-hour Azure network collapse spilling straight into Microsoft 365 in July, followed by an August authentication blunder that broke Exchange, Teams, SharePoint and Defender for days.
September to November: Fresh Azure gateway failures across late September and October, a Global Secure Access hiccup in mid-October, and finally November's nearly 24-hour global blackout across 74 countries, forcing UK hospitals to reschedule procedures, stalling court judgments and leaving jails, police and the US House of Representatives locked out of email.
Microsoft has plenty of company in this circus, from Google Cloud's October power fiasco in London freezing Revolut card payments to AWS outages and last week's parade of broken banks, airlines and spy networks. Still, nobody else has turned blowing up their own cloud into a reliable monthly subscription quite like Redmond. (Control Plane, The Guardian)
Now get a load of how Microsoft actually broke the world last week. In plain English, its own preliminary report blamed a hidden bug baked into a previous fix it shipped back in July 2026 for the networking gateways sitting in front of Microsoft 365. Under heavy traffic on newer servers packed with processor cores, internal processes began fighting over the same shared resource until more than 80% of the gateway fleet locked up. Because that dodgy fix had been marinating in live servers for over three months, Microsoft's usual undo buttons and feature flags could not touch it.
Even better, why did Microsoft's load testing completely miss this ticking bomb? Because its test labs were running on older, weaker hardware with fewer CPU cores than the real machines serving customers! Imagine being a multi-trillion-dollar company hurling absurd fortunes at AI hype while pinching pennies on the test servers keeping hospital email alive. (The Register, Neowin)
Of course, Microsoft's PR squad sprinted out mid-outage to swear up and down that the crash had nothing to do with "cyber or AI / agents". Fair enough, nobody outside Redmond has the commit log to prove Copilot personally typed that July defect. Look at the timing, though! CEO Satya Nadella had already bragged that 20% to 30% of the code in some internal repositories was written by software, and in July 2026, the exact same month Microsoft started pushing that broken gateway fix, he boasted to investors that one in three pull requests on GitHub now involves an AI agent.
And the cherry on top of the whole farce? On Monday morning, right before Microsoft Ignite kicked off, who got sent out onto Twitter to publicly grovel to locked-out hospitals, courts and businesses around the globe? Charles Lamanna, Microsoft's Executive Vice President of Copilot, Agents and Platform! You literally cannot script satire that sharp. Even if Copilot never laid a finger on that July patch, making the head of Copilot and Agents apologise on X because Microsoft's own under-tested fix bricked global email is pure poetic justice. (The Register, The Guardian)
Welcome to the AI ouroboros:
Step back from Microsoft and the real punchline of 2026 snaps into focus. Generative AI has turned modern tech infrastructure into an AI ouroboros, a self-feeding perpetual motion machine of crap code and corporate panic:
AI coding bots pump out bloated, plausible-looking code faster than humans can understand it > AI security scanners uncover bugs across legacy and new systems at breakneck speed > panicked IT teams shove massive, poorly tested patch dumps into live servers before AI-powered attackers exploit them > production explodes > depleted engineering teams reach for more AI to dig themselves out of the rubble.
This loop is spinning right out in the open. Publicly disclosed software vulnerabilities tracked in the National Vulnerability Database this year have already more than doubled all of 2025, with over half surfacing since August. Microsoft alone has patched 5,535 security flaws in its own products in 2026, more than double 2024 and 2025 combined, including a colossal November Patch Tuesday dump of 1,094 vulnerabilities that Microsoft proudly credited to AI just two days before Outlook died! When the NSA's classified internal network went down this week, Director of National Intelligence Jay Clayton tied it to patching PHP flaws uncovered with AI help, while Senate Intelligence Vice Chair Mark Warner spelt out the glaring bind: AI is exposing security holes so fast that organisations are ramming urgent fixes into older systems never built to survive that pace. (CyberScoop, PBS News)
Meanwhile, the executive delusion sitting on top of this inferno borders on clinical. A June 2026 New Relic survey of 200 US tech leaders found 94% rated AI-generated code higher quality than human code during review, right before admitting that 82% had suffered a production failure tied to AI code in the previous six months! Research from Cisco's Splunk division and Oxford Economics calculated that unplanned downtime now drains $600 billion a year from Global 2000 firms, up 50% in two years. Best stat of the whole circus? Every single surveyed tech leader admitted experiencing AI-related downtime, yet 56% of those exact same executives still claimed AI reduces downtime risk!
Of course they keep spinning the wheel! The exact same hyperscalers flogging the AI coding bots that bloat codebases are selling the AI security tools to scan the mess, while billing by the second for the cloud compute required to keep both sides of the fight running. Creating an infinite crisis that can only be treated by buying more of your own tokens is a racket most mob bosses could only dream of. (New Relic, Cisco Newsroom)
Amazon, rogue bots and the rest of the 2026 clown car:
If you think Microsoft is having a uniquely cursed year, take a look at Amazon. Late last year, two senior vice presidents co-signed an internal memo mandating 80% weekly adoption of Amazon's in-house AI coding tool Kiro, while the Stores division crowed about deploying 21,000 AI agents for 4.5x developer velocity and billions in savings. Soon afterwards, Kiro reportedly decided to delete and recreate a live AWS Cost Explorer environment, triggering a 13-hour outage.
Then March hit, bringing a string of retail meltdowns capped by a six-hour Amazon.com checkout blackout linked in reporting to an estimated 6.3 million lost orders! Amazon pushed back indignantly, insisting none of the outages involved AI-written code and only one involved AI at all, because an engineer acted on inaccurate advice an AI assistant inferred from an outdated internal wiki. Somehow Seattle thought "our bot didn't write bad code, it just hallucinated terrible instructions and nobody caught it" sounded reassuring. Either way, the exact same SVP who co-signed the Kiro mandate had to call an emergency meeting and impose a 90-day "code safety reset" across 335 critical retail systems, dragging humans back into the loop with what Amazon delicately called "controlled friction". (Business Insider)
And down in the rest of the industry, 2026 has read like a comedy sketch written by someone who hates computers:
OpenAI gets sued over its own escape artists: On Monday, California Attorney General Rob Bonta sued OpenAI over what the state alleges are more than 80 "loss of control" incidents, headlined by a July breach where testing agents broke out of an inadequately secured sandbox, hopped onto the live internet, pinched credentials and hacked Hugging Face! Even Bonta waved away the sci-fi mystique, calling it the predictable result of a company prioritising speed and hype over basic containment. (California DOJ, OpenAI)
Gone in nine seconds (with a polite apology): In April, a Cursor coding agent running Claude Opus 4.6 hit a minor credential mismatch at startup PocketOS, went rooting through an unrelated file for an overly broad API key, and deleted the company's entire live production database and volume backups in nine seconds. When confronted, it calmly wrote out a chat confession admitting, "I violated every principle I was given." Turns out typing "please don't nuke prod" into a prompt is not an actual security barrier! (The Register, Tom's Hardware)
The terraform destroy special: Two months earlier, a Claude Code agent ran terraform destroy against an outdated state file for DataTalks.Club, vaporising two and a half years of live student data, servers and automated snapshots until AWS support miraculously fished out a hidden backup. (AI Incident Database)
Vibe-coded security theatre: Viral AI-agent social network Moltbook, whose founder proudly boasted he hadn't written a single line of code himself, forgot basic database access controls and left 1.5 million API tokens and 35,000 emails sitting wide open for Wiz researchers. Not to be outdone, $6.6 billion vibe-coding darling Lovable left an API permissions flaw open on older projects for 48 days, letting any free account grab other people's source code, chat logs and database keys, then initially tried gaslighting everyone by calling it "intentional behavior" before backtracking! (Wiz, The Register)
Verdict
Here is the funniest and bleakest part of the whole saga: we are still only in November! Next month is December, when panicked middle managers crack the whip to hit end-of-year targets, push unfinished features before the holiday freeze and tick off quarterly AI adoption KPIs before disappearing until January. Just imagine the mountain of half-baked, AI-slapped time bombs about to be quietly buried in live servers next month, waiting patiently to detonate all over 2027! Even Cloudflare CEO Matthew Prince looked at this tidal wave of AI-discovered flaws and panic-patching back in October and warned Reuters that "we're going to have a couple years where every single week it's going to feel like the sky is falling."
Man, like I have talked about on this site before, it is genuinely baffling watching these trillion-dollar giants dig themselves deeper into the pit every single month. Honestly, the whole thing would be absolutely hilarious if it weren't actively wrecking real-world infrastructure and screwing with ordinary people's lives, cancelling hospital procedures, stalling courts, grounding flights and locking people out of their own money.
Yet these executives keep waltzing onto keynote stages with that shameless Todd Howard grin, swearing straight to our faces that "it just works™". After all, you have to keep selling shiny new fixes for the exact disasters you just caused. None of this is some unavoidable growing pain on the road to a glorious technological future; it is literally shit these companies manufactured themselves so they can keep pumping the AI bubble dream and charging everyone twice for the privilege of watching it burn.