Your cloud vendors are shipping AI-generated code. More outages are coming. | Tech Radar

Overview

News, deals, reviews, guides and more on the newest computing gadgets

Start exploring exclusive deals, expert advice and more

Details

Unlock and manage exclusive Techradar member rewards.

Unlock instant access to exclusive member features.

Get full access to premium articles, exclusive features and a growing list of member rewards.

Your cloud vendors are shipping AI-generated code. More outages are coming.

AI-generated code is making cloud outages more frequent and severe

When you purchase through links on our site, we may earn an affiliate commission. Here’s how it works.

Q4 2025 was a brutal quarter to be an enterprise IT leader. In October alone, AWS suffered a 15-hour DNS cascading failure that took down 141 services and impacted over 3,500 companies across 60 countries — Snapchat, Roblox, Fortnite, and airline reservation systems among them.

Azure followed days later with a networking configuration failure in its East US2 region that dragged on for nearly 50 hours. Cloudflare went down in November from a software defect triggered by a single database permissions change. Tens of thousands of Saa S outages occurred in 2025 alone.

Chief Product Officer and Co-Founder at APIContext.

Enterprise CIOs who rely on AWS, GCP, Azure, and the dozens of cloud vendors layered beneath and around them should not treat these as anomalies. They should treat them as a preview.

AI code security risk: The need for a smarter layer between detection and remediation

Tame your AI gremlins before the chaos becomes permanent

Because the recent conversation inside Amazon about how fast engineers should be shipping AI-generated code is a canary in the coal mine for every enterprise that runs critical workloads on infrastructure someone else builds.

Amazon is not alone. Across the technology industry, the adoption of AI coding tools has moved from experiment to expectation at extraordinary speed. Among U. S.-based developers, 92% report daily use of AI coding assistants.

Among Fortune 500 companies, nearly all have adopted at least one vibe coding platform. Google has disclosed that more than 25% of its new code is now AI-assisted.

That last number is worth pausing on: a meaningful fraction of the code running Google's cloud infrastructure that your enterprise depends on was not fully written by an engineer who read every line.

For startups building internal tooling, maybe this is mostly fine. The blast radius of a bug is contained. For the hyperscalers that underpin global enterprise IT, it's a different calculus entirely.

The AWS moment of truth is significant not because Amazon is uniquely reckless, but because it's the first time the internal pressure to ship faster using AI has leaked out of a major cloud provider in a documented, reported way. Every major cloud and B2B software vendor is navigating this tension.

The challenge is that when these cloud services fail, it becomes your operational failures, often with no warning and sometimes without even a timely status update.

You can’t firewall a conversation: how AI red-teaming became mission-critical

Tech leaders need a cloud reality check before it’s too late

The confidence problem hidden in AI-generated code

Understanding why this matters requires grasping one non-obvious characteristic of how AI coding tools fail. Large language models produce code with syntactically uniform confidence. They write a critical distributed locking function with the same assurance they write a sorting utility.

The code looks correct. It often passes tests. The failure surfaces under specific timing conditions, specific load profiles, specific combinations of infrastructure state that nobody thought to write test cases for, and that the model certainly didn't flag.

Security researchers have documented that AI-generated code exhibits significantly higher rates of common vulnerability classes (buffer overflows, race conditions, improper input validation) compared to hand-authored code — not because models are careless, but because they learned from thirty years of accumulated human mistakes in their training data.

The Cloudflare outage in November 2025, where a duplicate entry in a bot management file caused cascading failures, illustrated the underlying failure mode perfectly.

While this was not categorized as an vibe coding issue, the change was implemented without adequate coverage of the specific runtime conditions that mattered. The operational consequences were global. AI code makes this exact failure pattern much easier to repeat, at higher frequency, across more vendors simultaneously.

The data from measuring API reliability across cloud providers in the past two years is unambiguous: things are getting worse.

In 2022, 18% of cloud services achieved 99.99% uptime. By 2023, that number had fallen to 7%. In our most recent analysis of 27 cloud services, not one achieved five nines — the historic telecoms standard for availability.

Our research across nearly 10,000 API endpoints and one billion API calls estimates that poor API quality now costs organizations billions in wasted developer effort alone, before you count the downstream business impact of actual outages.

Third-party monitoring data corroborates the deterioration. Average weekly API downtime increased 60% between Q1 2024 and Q1 2025 — from 34 minutes per week to 55 minutes per week. Average API uptime fell from 99.66% to 99.46%.

Those numbers are small on paper. In practice, a 0.2 point drop in uptime across dozens of cloud dependencies compounds quickly for enterprises running complex, multi-vendor architectures.

An industry that is shipping more code, faster, with less human engagement per line written, while maintaining the same (or reduced) investment in chaos engineering and fault injection testing, is going to produce more production failures. That is what the data suggests is happening.

Your vendor's status page is not your early warning system

This is the part of the conversation that matters most for enterprise CIOs, and it's the part most often skipped in vendor relationships.

When the AWS October 2025 DNS failure hit, more than 4 million outage reports were submitted by users within the first two hours.

The companies that knew about it earliest were not reading AWS's status dashboard — they were already monitoring their critical API paths from independent vantage points and had alerts firing before AWS had officially acknowledged the incident scope.

Azure’s outage in October 2025 followed the same pattern, users couldn’t report issues because the support portal to report them was affected by the outage. Vendor status pages consistently lag the actual event by meaningful intervals.

The fundamental problem is that most provider status pages require human intervention to update. In the middle of a major incident, the engineers who would update the status page are triaging the incident. You find out when they find time to tell you, not when the problem starts.

Which means that for enterprise IT teams whose SLAs, customer commitments, and incident response plans depend on knowing about outages quickly, relying on vendor status pages is an operational gap that is going to hurt more as the frequency of outages increases.

So, more AI-assisted code is being shipped by your cloud vendors, your software vendors, and increasingly your own internal teams.

The testing and validation practices that would catch AI-generated errors before they reach production are not scaling at the same rate as development velocity. The Q4 2025 outage cluster is not an aberration — it is a leading indicator.

The appropriate response for enterprise IT leaders is not to demand that vendors stop using AI tools. That ship has sailed, the economics are too compelling, and frankly some of this code is genuinely good.

The appropriate response is to treat your cloud vendor relationships the way a mature security team treats software supply chains: with the assumption that something will eventually go wrong, and with the infrastructure in place to detect it independently before it cascades. Concretely, that means:

Independent API monitoring that runs from your users' vantage points, not from your vendors' data centers. When a cloud provider's DNS layer fails, their own internal monitoring often fails alongside it — as it did for AWS in October 2025. External monitoring from diverse geographic vantage points catches what vendor dashboards miss.

Real-time baseline visibility across all cloud dependencies, not just your primary provider. Enterprise architectures now span AWS, Azure, GCP, and dozens of Saa S vendors with their own infrastructure dependencies.

A failure anywhere in that chain can propagate unpredictably. You need to see the whole delivery chain, not just the tier-one relationships.

Shorter alert latency with automated triage, not manual monitoring. The value of knowing about an outage ten minutes after it starts versus sixty minutes after it starts is enormous — the difference between proactive customer communication and reactive damage control.

The moment of truth conversation happening at AWS is, ultimately, a conversation happening across the entire industry about how to maintain quality as AI accelerates velocity. Enterprise IT leaders don't get to wait for that conversation to conclude.

Your vendors will figure it out eventually. In the meantime, your systems will be the ones absorbing the variance.

More outages are coming. The question is whether you see them before your customers do.

This article was produced as part of Tech Radar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of Tech Radar Pro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

Chief Product Officer and Co-Founder at APIContext.

You must confirm your public display name before commenting

1 Irked by Air Pods Max 2's 6-year-old design? Apple's cheaper Beats arm has just the thing, as new over-ears are spotted at the FIFA World Cup training camp

2 Windows 11 is finally getting the Start menu changes we all wanted — and a surprise bonus

3 What is the release date for Dutton Ranch episode 5 on Paramount+?

4 Oops! A Samsung employee may have just leaked the Galaxy Z Fold 8 — here's how it compares to the narrower Galaxy Z Fold 8 Ultra

5'Welcome to the family' — Qualcomm says it's glad Nvidia is joining the Arm race with the RTX Spark

Tech Radar is part of Future US Inc, an international media group and leading digital publisher. Visit our corporate site.

Key Takeaways

News, deals, reviews, guides and more on the newest computing gadgets
Start exploring exclusive deals, expert advice and more
Unlock and manage exclusive Techradar member rewards
Unlock instant access to exclusive member features
Get full access to premium articles, exclusive features and a growing list of member rewards