AI Discovers Vulnerabilities in Software Born 38 Years Ago: A Rush of Fixes Begins Behind the Scenes of the Internet

AI Discovers Vulnerabilities in Software Born 38 Years Ago: A Rush of Fixes Begins Behind the Scenes of the Internet

The money that should have been transferred is not reaching the recipient's account. In such cases, the first thing you want to know is whether you should proceed with the process again. However, when you open your smartphone to check the situation, you not only see the bank malfunction but also news that emails are down and flights are grounded. Even if each is a separate event, when they appear on the same screen, it seems like something major is happening.

In November 2026, such days actually continued in the United States. Major bank JP Morgan Chase faced transaction processing issues. Microsoft and Apple's services also experienced successive malfunctions, and domestic flights in the U.S. were temporarily halted.

According to the New York Times on November 21, U.S. authorities explained that they have not confirmed any evidence indicating that the series of disruptions were caused by malicious cyberattacks. For some of the disruptions, excluding the nationwide flight halts, officials pointed out that it was rather the work of those trying to protect the system. It is possible that the software update intended to patch vulnerabilities did not go well.


Behind the scenes of opening the usual file, software from 38 years ago is at work

In considering the relationship between the recent disruptions and AI, the paper introduced software called "Ghostscript." Responsible for document and image conversion, its inception dates back 38 years. Even if users are unaware of its name, files entrusted to web services are processed by such software behind the scenes.

Security researcher Paulos Ibero used AI to investigate what happens inside the service that receives files and found a significant vulnerability in Ghostscript. It was an issue that could affect the services of multiple companies. Ibero reported it to the developers and related companies, and corrections were made.

The "38 years" refers to the length of the software's history. It does not mean that the vulnerability remained unchanged for 38 years. What should be noted is that many companies relied on a single piece of software that is usually unnoticed. The discovery by one researcher led each company to check where it might affect them and to deliver necessary corrections.


Even if AI speeds up discoveries, testing updates takes time

For example, suppose a service that accepts documents updates its document conversion software. Even if the developer confirms that the vulnerability has been patched, the company has another confirmation to make. Can the application forms that trading partners have been using for years still be processed as before? Can a large volume of documents at the end of the month be processed without delays? These are things that can only be confirmed by those who know the actual operations.

Once a vulnerability is identified, you want to fix it quickly. On the other hand, to minimize the impact on users, time is needed to apply the update partially and observe the results. The more vulnerabilities AI finds in a short time, the heavier the judgment becomes on which to fix first and how much to confirm before applying widely.

Reuters also reported on November 18 that the increase in correction work might be related to some of the disruptions in the U.S. What should be distinguished here is that the Ghostscript case is an example of vulnerability discovery by AI and not presented as the cause of Chase's transaction delays or flight halts. The intersection of AI and disruptions lies in how the work of maintaining software changes, separate from the causes of individual incidents.


On social media, a "single explanation" linking banks and airlines spread

According to the U.S. news site Axios, when the Federal Aviation Administration halted flights, a view emerged on social media linking it to the introduction of AI tools at the agency. There were also posts linking Microsoft's disruptions to the company's personnel cuts. Both were speculations based on the chronological order of events.

Among them, reactions regarding banks were particularly eye-catching. In a survey by social media analysis company PeakMetrics, more than half of the posts about the Chase disruption also mentioned other disruptions, such as flight halts. Furthermore, the company estimated that about 30% of the posts spreading conspiracy theories about the six disruptions surveyed were from accounts believed to be bots performing automatic postings. This approximately 30% differs from the proportion of people who wrote impressions or inconveniences about the disruptions.

A person waiting for a deposit writes, "It hasn't arrived yet." That post conveys what users are troubled by. On the other hand, a post saying "both the bank and the airport are under the same attack" requires separate evidence linking the two sites. On the timeline, these two types of information are displayed in exactly the same size.


"10:50 AM" conveyed by Japanese banks after a transfer disruption

There are clues to consider from the user's perspective in Japan as well. From October 10 to 11, 2023, a malfunction occurred in the Zengin System, which connects financial institutions. Some banks were unable to make transfers to other banks, and there were delays in deposits. Mitsubishi UFJ Bank later accepted inquiries about transfer and deposit delays and also compensation consultations from users who incurred additional costs due to the disruption.

On the following day, the Zengin Network announced that as of 10:50 AM, the processing between the Zengin Center and banks for transfer transactions that could not be handled in the previous two days was completed. This is much more specific than just saying "restored." For users, there is a significant difference between being able to accept new transfers and having already processed the transfers that were previously received. Furthermore, whether the deposit has been credited to one's account remains an individual matter to be confirmed.

What the series of disruptions in the U.S. also poses to Japan is the question of what can be communicated while it takes time to investigate the causes. Is it necessary to resend the transfer? How much of what was accepted yesterday has been processed? When will the next update be provided? No matter how much AI speeds up the discovery of vulnerabilities, what helps those waiting for deposits is information that allows them to judge the status of their specific case.


Source