Are we toast?

Written by 5ivecanons Creative Staff

It’s been a hard week for the systems everyone counts on. On November 12, Microsoft 365 went down worldwide, taking Outlook, Teams, and SharePoint with it for the better part of a day and leaving services shaky into a third. The next afternoon, the FAA issued a nationwide ground stop after air traffic control centers began losing communications. It lasted about four and a half hours and led to nearly 8,000 cancellations and more than 10,000 delays. More than 17,000 flights were canceled before the weekend was over, and American Airlines, the hardest-hit carrier, was still working through cancellations on Monday while outage reports for its website spiked that evening. That same Monday, a software bug at JPMorgan Chase slowed account access and payments, and customers at Bank of America, Wells Fargo, and U.S. Bank felt it through delayed Zelle transfers.

Separate incidents, separate causes, and by every official account, no attack behind any of them. But they arrived against a backdrop of AI, newly discovered software vulnerabilities, and relentless bot traffic that makes the internet feel harder and harder to keep running. How much more pressure can the whole thing take?

We build websites, manage digital platforms, and help businesses communicate. Increasingly, that work includes figuring out why a website is serving an endless stream of machines while an actual customer waits for a page to load.

That’s a frustrating conversation to have with a client. They paid for a website. They pay for hosting. They invest in maintenance. They reasonably expect someone to be able to visit it.

Meanwhile, somewhere, a bot is asking for everything. Again.

How do you find the truth, when AI is doing the searching?

Source What it found
Imperva, 2026 Bad Bot Report Automated traffic was more than 53% of web traffic in its 2025 analysis, up from 51% in 2024.
Fastly, 2025 Threat Insights Report Bots were 37% of observed traffic. Of that bot traffic, 89% was classified as unwanted.
Cloudflare, 2025 Year in Review AI bots averaged 4.2% of HTML requests across its customer base, with Googlebot measured separately. Crawling triggered by people using AI assistants grew more than fifteenfold during the year.
Cloudflare Radar, 2026 Automated traffic crossed 57% of HTML requests in June 2026, and by mid-year roughly half of crawler requests were for AI training, up from about 22% in spring 2025.

These reports cover different networks, periods, and definitions, so their percentages measure different things. The direction is the same in all of them, though. A substantial share of the activity reaching websites comes from software, and the trade is uneven. Cloudflare’s crawl-to-refer data from this year shows traditional search fetching roughly five pages for every visitor it sends back. AI training crawlers fetch hundreds or even thousands of pages per referral.

Some of that software is useful. Search engines need to index pages. Monitoring tools check whether a site is working. People increasingly ask AI assistants to research products and services on their behalf. Other bots scrape content, test stolen passwords, probe for vulnerabilities, or collect whatever they can reach. And even a crawler with a legitimate purpose can create problems if it requests too much, too quickly.

A server has to handle the request before anyone benefits from the visit. Machines can consume a business’s resources without becoming customers, filling out a form, or contributing anything toward the bill.

AI is changing more than the volume of crawling, too.

On October 21, 2026, the PHP project released emergency security updates for two critical vulnerabilities known collectively as PHP2Shell, both rated 9.2 out of 10 in severity. PHP is part of the underlying software that runs WordPress and many other web applications, and the flaws could allow attackers to run code on vulnerable servers without logging in. The distinction matters. This was a problem beneath WordPress, not in it, so updating a theme or plugin alone wouldn’t resolve it. Exposure depended on PHP version, configuration, and application behavior, and it didn’t mean every WordPress website had been compromised. WordPress.com, for example, said its hosted sites were already protected.

The discovery is the part worth sitting with. The flaws had been sitting in PHP for roughly a decade, and they were found twice in one season, independently, by research teams using AI tooling. That’s good news for defenders and sobering for everyone responsible for keeping software secure, because the same advances that help researchers find old flaws also lower the effort required to understand and exploit them. Security firms reported exploitation attempts beginning days before the patches shipped, and attackers have since chained the flaws into real intrusions.

A website doesn’t have to be hacked to become unusable, though.

We’ve been working through recurring traffic problems on a client website. We’ll keep the client anonymous, because this experience could belong to plenty of businesses.

The pattern is exhausting. Automated requests keep coming. Blocking individual IP addresses provides limited relief when traffic can arrive through changing addresses. Crawlers work through pages, and the server spends resources answering requests that have little apparent value to the business. During one support review, the hosting team shared counts totaling more than 1.5 million requests across five browser identification strings. Those were request counts, not unique visitors, and the excerpt didn’t establish a reporting period. But they helped show the scale of what we were investigating.

For requests that can’t be served from a stored copy of a page, WordPress relies on PHP workers, the finite pool of processes that handle server-side work. When too many uncached requests arrive together, work backs up and visitors see delays or errors.

So we worked with the hosting team to filter the traffic. Then came the second problem: protecting access without blocking the people who need it. A rule targeting traffic identifying itself as older browsers was too broad, and legitimate visitors, including our own team, started receiving access errors. The rule needed adjustment. An additional protective mode introduced a browser challenge, and we found that the site’s customer relationship platform connection needed attention too. We couldn’t establish that the protective mode caused that disconnection, but we checked, because a working homepage doesn’t mean much if the forms behind it are quietly failing.

That’s what maintenance looks like now. Investigating traffic, adjusting protections, testing the actual customer experience, and repeating.

Even the companies that filter the internet’s traffic for a living have felt this. On November 18, 2025, Cloudflare suffered its worst outage since 2019, and the trigger was its own bot defense. A routine database change caused the configuration file its Bot Management module reads to double in size, the oversized file tripped a hard-coded limit, and the software routing traffic for millions of websites crashed. A month earlier, a race condition in Amazon’s DNS automation left a critical database endpoint with an empty address record, and roughly 15 hours of cascading failures followed across 142 AWS services. The machinery built to manage machine traffic has become so central that when it stumbles, everyone falls with it.

For a company website, an interruption means lost inquiries, delayed orders, and damaged trust. Extend the dependency to the services that keep communities functioning, and the stakes stop being abstract.

In healthcare, unavailable systems delay eligibility checks, prescriptions, payments, and care. After the 2024 Change Healthcare cyberattack, an American Hospital Association survey of nearly 1,000 hospitals found 74% reporting direct patient care impacts and 94% reporting financial impacts. In banking, a payment delay becomes a payroll problem. Monday’s Chase disruption showed how trouble at one institution slows transactions for customers of banks whose own systems were working fine. In manufacturing, the September 2025 cyberattack on Jaguar Land Rover halted Britain’s largest automaker for roughly six weeks, cost the company £196 million in a single quarter, and cost the UK economy an estimated £1.9 billion across more than 5,000 organizations. Behind each of those numbers are patients waiting for care, workers waiting for pay, students locked out of lessons, restaurants unable to ring up orders, and shipments waiting for instructions.

The causes deserve precision. Microsoft traced its outage to a problem introduced during a network capacity upgrade; servers began failing automated health checks and dropping out, overloading the ones that remained. The FAA’s ground stop traced to a communications failure during a long-running program to replace copper lines with fiber, and officials said they found no evidence of anything malicious or AI-related. Chase blamed a software bug in on-premise hardware and said the issue wasn’t cyber-related. A software defect, an overloaded application, and a deliberate attack require different responses. They just leave people facing the same immediate problem: something essential no longer works.

For businesses, that makes maintenance and continuity part of operating responsibly rather than an add-on. Someone has to know who manages the underlying software, whether backups actually restore, whether forms and payments complete, and what the automated traffic is doing. Security changes need testing against the ways customers actually use a site, and there should be a way to reach people when the normal platform is down. Those responsibilities belong in budgets, project scopes, and honest conversations about what a digital platform costs to operate.

We have to account for that too. Every tool we recommend adds a dependency. Every protective measure needs testing. Every customer being asked to try again is a person whose time matters.

Are we toast? We still have choices about what we build, what we maintain, and how we prepare. Making those choices takes more attention than it used to, and the attention is the point. These systems stay available because people keep them available. The internet is getting busier with machines. Keeping it useful for people is becoming a bigger part of the job.