A string of major outages hit big tech, financial institutions and critical national IT infrastructure last week, sending Americans into fear. After the Federal Aviation Administration (FAA) ordered a nationwide stop to aircraft takeoffs on Friday the 13th local time, social media users voiced concern about possible hacking and the risk of bigger incidents to come.
Downdetector, a site that tracks internet outage reports, received nearly 6 million reports of problems in the United States alone over the week through the 16th.
Axios said it was the highest weekly figure in the past year. In a separate development, Homeland Security Secretary Markwayne Mullin went so far as to mention that AI systems that write and fix code could have affected the outages.
With attacks using AI hacking tools pouring out in October to the point the period was dubbed “Black October,” the concerns surfacing in many places sounded plausible.
In the end, the concerns remained concerns for now. Microsoft, the FAA, Chase bank, the National Security Agency (NSA), Apple and United Airlines all made clear the outages were not due to hacking. It was not the case that AI slipped out of control and brought down systems. Most were cases in which companies and institutions ran into incidents while upgrading systems.
The Washington Post reported that government and bank officials told lawmakers the outages were not due to a cyberattack but occurred when patches applied to fix security vulnerabilities found by AI models temporarily overloaded systems.
Axios, citing experts it interviewed, said the recent outages were “far from hacking or a cyber catastrophe.” It said most were problems during software patching, and links between the incidents were also hard to find.
Axios said Chase said transactions with other banks were delayed due to an internal software bug, while United said only that it was a technical problem. The FAA explained that a rush of data across its network overwhelmed aging equipment. Microsoft, for its part, suffered an outage due to problems that arose during an earlier network capacity upgrade.
Even if recent outages were not caused by hacking or AI, they are not something that can be called entirely unrelated to AI.
Some see the series of IT system failures as a scene symbolising the rise of so-called “Bugmageddon,” as AI-powered cyberattacks continue and software security patches pour out in response.
The Wall Street Journal (WSJ) reported that “IT staff are losing sleep, and incident-response specialists are turning down even new requests because there is too much cybersecurity work.” It said service outages were not new, but AI has made the situation more complex, making it harder to distinguish between cybersecurity threats and constant system updates made to address them.
The Washington Post also cited a congressional official who received briefings from government and bank officials on the incidents as saying the pressure to patch immediately has grown as AI models find large-scale vulnerabilities. The official said more patches are being pushed at once than systems can handle.
Anne Neuberger (앤 뉴버거), a former White House deputy national security adviser and a general partner at Silicon Valley venture capital firm Andreessen Horowitz (a16z), said there are two major problems. One is systems known for a long time to be aging, like the FAA system, that need major upgrades. She said AI-based attackers capable of finding code security vulnerabilities have also emerged.
In a recent WSJ op-ed, she also said AI is reducing the time from vulnerability discovery to attack. “If you defend at human speed, you lose the race,” she wrote, adding that defences for banks, hospitals, and infrastructure such as power and water must be changed quickly.
Ian Mulholland (이언 멀홀랜드), chief information security officer at Salesforce, told the WSJ that so many reports have warned that an AI apocalypse is imminent that when ordinary people see systems stop working, they immediately think it is due to a cyberattack even when it is not.
As AI finds vulnerabilities quickly, the burden and risk associated with fixing them and deploying patches are growing.
Patch work may seem like something that can be done simply by moving fast, but in most cases it is not. Patches can be made quickly, but deploying them to a system can be a different issue. Because problems can occur during deployment, IT staff have little choice but to weigh various factors before rollout. Many of the recent IT outages also occurred during patching. The WSJ, citing cybersecurity experts, said the worst outcome a patch can cause is a service outage.
Axios also reported that the spread of AI-assisted coding is another variable. It said AI has greatly accelerated how fast engineers build and refine software, but bottlenecks have emerged in the stage of reviewing whether new code is safe and reliable. It said mistakes can follow if review staff or systems cannot handle the workload.
David Woods (데이비드 우즈), emeritus professor at Ohio State University in integrated systems engineering, said complex systems are being pushed to run faster and cheaper without enough consideration of how they might fail. He said cross-checks, stress tests and safeguards are being removed one by one.
Some also question the blame placed on AI itself. Liz Fong-Jones (리즈 퐁존스), a technical fellow at Honeycomb, said there is little in common across these incidents. She said the outages being seen now are simply cascading failures, a phenomenon decades old, and have nothing to do with writing and deploying code with AI.