i
News
News · 2026-11-21

Technology outages are accelerating alongside vulnerability discovery

Dataist Dataist

Large-scale outages are only the beginning

Cover: Technology outages are accelerating alongside vulnerability discovery

Outages affecting Outlook, Apple services, six US banks, United Airlines and the NSA’s closed network were not linked to a coordinated cyberattack; each had an ordinary technical explanation. But researchers warn that AI is speeding up the discovery and exploitation of vulnerabilities, forcing organizations to test and install updates faster. In 2026, the number of vulnerabilities discovered exceeded 100,000, up from about 25,000 in 2022, before large language models became widespread. IT teams face a growing workload, and mistakes in updates can trigger new outages.

Technology outages are accelerating alongside vulnerability discovery

Updates need to be tested faster

Recent incidents include a major Microsoft 365 outage caused by a faulty infrastructure update and compounded by other network problems. This week, NSA officials said an internal communications system went down after an unsuccessful update installation following critical October PHP updates. Apple also linked problems this week to an internal update.

On Tuesday, US Senator Mark Warner said that AI-assisted vulnerability discovery and the subsequent installation of fixes are putting enormous strain on systems not designed to handle changes at this pace. Microsoft CEO Satya Nadella warned that instability could worsen: if attackers are already exploiting a vulnerability, a released fix needs to be installed in production immediately.

External pressure and update errors are not new. Another problem is the “technical debt” many organizations have accumulated: for years, they have skimped on upgrading and maintaining digital systems. Outdated equipment and inadequate safeguards have long posed risks, but organizations could put off updates and often avoided serious consequences.

Joe Slowik, Dataminr’s director of threat research and cyber engineering, says companies once typically had time to test and install a fix after a serious vulnerability was patched. Now, the window can be so short that there is no time left for thorough testing.

IT administrators are also using AI to assess updates faster and prepare systems for installation. But today’s tools do not replace human analysis and planning. Lea Schönherr, a lecturer at the CISPA Helmholtz Center for Information Security, Germany’s national research institute, says specialists still need to determine what caused an outage, make sure the fix addresses that cause, and check systems for similar problems. Installing updates across widely used systems also takes time.

Cloudflare CEO and co-founder Matthew Prince predicts the next two years will feel like a catastrophe. At the same time, he expects AI-driven acceleration to eventually help security specialists handle common vulnerabilities more effectively and improve software security overall. In his forecast, new threats and vulnerabilities will arise less often, and the overall flow of problems will decline substantially.

More vulnerabilities, but not all matter

In 2026, the total number of vulnerabilities discovered exceeded 100,000. For comparison, about 25,000 were recorded in 2022, the last year before large language models became widespread. With more than a month left in the current year, the figure had already reached twice the combined number of vulnerabilities recorded in 2024 and 2025.

Over the past few days, researchers reported using AI to find critical vulnerabilities in PHP — for the second time — as well as in the open-source NGINX web server and the Java programming language. Forbes also reported 35 vulnerabilities in iPhones, Macs, iPads, Apple Watches and other Apple products that researchers found with AI.

But the number of vulnerabilities discovered is not the same as the number that pose a serious threat. Slowik says the number of issues found has risen sharply, while the number of vulnerabilities people really need to worry about has stayed roughly the same. At the same time, attackers have become much faster at turning vulnerabilities into working attack tools.

This year, only 300 entries were added to the US Cybersecurity and Infrastructure Security Agency’s catalog of exploited vulnerabilities. That is far below the total number of flaws discovered. VulnCheck’s own database lists a higher number of known vulnerabilities already used by attackers: 1,500. Even that estimate is only 1.5% of all vulnerabilities discovered in 2026.

Schönherr urges people not to blame AI when things go wrong. In her view, responsibility lies with the people and companies that choose to use AI and give it access to infrastructure; those decisions should have consequences and be subject to rules.

Other news

Data from JPMorgan Chase, Visa and Mastercard. After outages earlier in the week, JPMorgan Chase confirmed that attackers stole some company data in a separate incident. The bank said the attackers accessed marketing information held by its contractors. According to a statement published by Reuters, the information “may” include customers’ names and email addresses. The company stressed that the incident was unrelated to the recent outages. Data belonging to Visa and Mastercard was also affected in attacks on their suppliers.
Attempts to attack Revolut customers. Dutch police said they had prevented several attempts to use violence against customers of digital bank Revolut. In September, attackers reportedly hacked the fintech company and stole customer data. The national police said several arrests had been made in connection with the theft. It is unclear who the attackers intended to target. Physical attacks have become more common in recent years. Sometimes called “wrench attacks,” they involve kidnapping and holding wealthy cryptocurrency owners hostage to force them to hand over money.
The UK and access to iCloud data. Following a court decision, the British government acknowledged that it had required Apple to bypass its own encryption and provide access to customers’ iCloud data. Previously, the authorities had argued that confirming the “technical capabilities notice” issued to Apple would harm national security. Under British law, such notices must remain secret. But, according to Computer Weekly, the order was widely known: senior Trump administration officials criticized it publicly, and British government representatives reportedly discussed it with newspapers.

Dataist's take: AI is speeding up both vulnerability discovery and exploitation, but that does not relieve organizations of responsibility for their infrastructure and the installation of fixes. The number of problems found alone does not show the scale of the immediate threat. What matters more is how quickly vulnerabilities start being exploited and whether companies have time to test updates before installing them.

Daily AI news

Every day we pick what actually matters in AI and explain it plainly — no hype, no filler. Subscribe if you want to follow where the industry is going.

Only what matters — every day

Follow on X