Back to Flash News

The New York Times: AI Is Speeding Up the Exposure of Old Code Vulnerabilities, and Technology Disruptions May Become the Norm

The New York Times reported on November 21 that artificial intelligence has significantly changed the cybersecurity landscape within months, accelerating the discovery of long-dormant software vulnerabilities and making serious technology disruptions potentially more common than before.

Security researcher Paulos Yibelo and his company PWN AI used AI to find the Upload2Shell vulnerability in Ghostscript image and document processing software developed 38 years ago. The vulnerability can allow attackers to achieve remote code execution on systems that make heavy use of the software, including Google and Zoom, by uploading files. Yibelo has reported it to developer Artifex and related companies, and the relevant patch has been released.

Over the past two weeks, multiple technology disruptions have occurred in the United States, including disruptions to JPMorgan Chase transaction processing, service outages at Microsoft and Apple, the NSA’s shutdown of critical systems, and the Federal Aviation Administration’s temporary ground stop on commercial flights. U.S. officials said there is currently no evidence that these incidents were caused by malicious cyberattacks, and some of the disruptions were related to companies and government agencies rushing to patch newly discovered vulnerabilities, with failures occurring in the patching process itself.

Original source:

  1. Title: Tech Outages Are Only the Beginning. A.I. Will Bring More Disruptions.
    Publisher: The New York Times (Dustin Volz, Kate Conger, Dylan Freedman, Sheera Frenkel)
    Publication date: 2026-11-21
    URL: nytimes.com/2026/11/21/us/politics/tech-outages-ai-disruptions.html
    Nature of source: media report
    Verification status: actually read the publicly available part of the article in this report (including the title, lead, the Paulos Yibelo case, and the description of recent disruptions; accessed on 2026-11-21, full paywalled content not read)
    Supporting content: confirms that AI is accelerating the discovery of old code vulnerabilities, the Upload2Shell case, recent multiple disruptions, and officials’ denial that they were attacks

ABAB AI Insight

Since Ghostscript was first released around 1988, it has long served as the underlying engine for PostScript and PDF processing and has been used by many applications, mail systems, print queues, and image processing tools. Upload2Shell is a use-after-free vulnerability in its Type 1 font loader, which can bypass the -dSAFER sandbox restriction and enable command execution. Paulos Yibelo previously conducted research at Amazon and now focuses on AI-assisted vulnerability hunting through PWN AI; this case shows that AI can systematically probe file-upload processing paths rather than relying on manual line-by-line review.

In terms of capital and business pathways, AI models are simultaneously playing the roles of discoverer and potential attack tool. Defenders are using AI to accelerate vulnerability identification and push patches, while attackers can also use similar models to shorten the window from discovery to exploitation. The recent disruptions did not result from confirmed malicious intrusions, but from operational failures during patch deployment, showing that the bottleneck of discovery speed outpacing repair and verification capability has begun to turn into real business disruptions.

Compared with major historical vulnerabilities such as Heartbleed or Log4Shell, the difference this time is that the discovery tools themselves are now automated and scalable. Heartbleed relied on manual or semi-automated scanning, whereas AI can now probe widely deployed legacy libraries along multiple paths in a short time. The similarity is that once a widely used underlying component is broken through, the impact area expands rapidly; the difference is that the cost of discovery has fallen sharply, while patch coordination and regression testing costs have not fallen in step.

This change directly raises operational risk for companies and government agencies that rely on legacy open-source components. A higher patch frequency itself may become a new source of failure, while demand increases for security engineers who understand old code context. If institutions continue to replace some development and operations roles with AI without at the same time retaining deep understanding of historical codebases, their ability to assess and prioritize newly discovered vulnerabilities may become further constrained. Metrics to watch include the patch coverage rate for Ghostscript and related image/document processing libraries, as well as the average remediation time of large institutions after high-risk vulnerabilities are disclosed.

ABAB News · Cognitive Law

  1. AI turns dormant vulnerabilities in old code from “possibly existing” into “batch-discoverable”

  2. When discovery speed exceeds patch verification speed, the hardening process itself can become a source of disruption

  3. Once AI breaks through a widely used underlying component, the scale of impact is determined by deployment breadth rather than attack complexity