This week, nearly everyone in the United States felt the impact of a series of significant network issues, resulting in flight cancellations, banking disruptions, and much more. Experts indicate that a critical factor behind these incidents is the speed at which artificial intelligence (AI) identifies software vulnerabilities, which has outpaced the ability of companies to patch these leaks. As a result, the potential for further system outages in the coming months looms large.
A source close to the core of the events recently shared insights, revealing that with significant advancements in AI capabilities, these technologies can detect a multitude of software flaws that humans may struggle to identify. To prevent hackers from exploiting these vulnerabilities, companies are hurriedly attempting to update their systems. Ironically, the simultaneous push for updates is overwhelming many legacy and complex computer systems.
Take Microsoft as an illustration: data from the cybersecurity firm Qualys shows that security updates released in November of last year addressed 63 vulnerabilities. In contrast, the updates in the same period this year remedied an astonishing 1,235 issues.
More than six months ago, both Anthropic and its competitor OpenAI issued alerts, indicating that their latest AI models could autonomously discover and exploit unknown vulnerabilities within the large systems of major corporations, even without explicit instructions. Subsequently, these corporations were permitted to utilize these models in controlled environments to detect and remedy bugs within their systems.
These powerful AI models can identify dozens of errors at once. However, it is crucial to distinguish between “discovery” and “fixing.” Many corporate systems are often cobbled together from various software developed over the past thirty years, making it essentially impossible to complete all repairs in one go. According to the informed source, this situation is leading to outages, service interruptions, and various other complications.
The source elaborated, stating, “These frontier models have crossed a crucial threshold in terms of cybersecurity capabilities, enabling them to find a tremendous number of vulnerabilities, which actually benefits security protection. However, the sheer volume of issues makes it difficult for companies to keep up with patching, presenting a significant challenge.”
“Nevertheless, we must remember that ultimately, patching vulnerabilities can make systems safer and less susceptible to attacks.”
As these models have been released in batches, the source anticipates that, with more companies adopting them, additional system disruptions may occur in the forthcoming months. To keep pace, the frequency of software updates must also increase.
This week, as Thanksgiving approached, tensions escalated significantly due to ongoing issues.
In recent days, multiple significant systems in the United States have encountered failures, affecting financial sectors, airlines, and government agencies.
– Microsoft: The Outlook email service experienced nearly 24 hours of disruption starting from November 12.
– Federal Aviation Administration (FAA): On November 13, the FAA ordered a nationwide halt to flights for over four hours, leaving many travelers stranded.
– JPMorgan Chase: On November 16, some customers faced difficulties with transfers and bank transactions.
– National Security Agency (NSA): On November 17, a malfunction occurred within its secure internal network.
Jay Clayton, the U.S. Director of National Intelligence, informed Fox Business that a newly discovered vulnerability in PHP—a programming language supporting most websites—required remediation, causing “some systems to fail to restore normal operations after applying the patch.”
This vulnerability was uncovered using a program developed by OpenAI, which utilizes AI to identify software bugs.
Microsoft, the FAA, and JPMorgan Chase attributed these issues to system failures, explicitly stating that they were not the result of a cyberattack.
Dario Amodei, CEO of Anthropic, suggested that this challenging adjustment period would ultimately yield benefits. He noted in May that defenders had a timeframe of six to twelve months to resolve all issues to prevent AI systems developed by the Chinese Communist Party from catching up or exploiting these vulnerabilities for attacks.
He remarked, “If we handle this correctly, we can find ourselves in a position better than we were initially because all these vulnerabilities have been patched.”
Cybersecurity expert Scott Schober also warned that fraudsters might target disrupted travelers, impersonating airlines to offer “flight rebooking” services. He advised, “If your flight experiences issues, please reach out directly to your airline.”
Clayton stated that there is currently no evidence suggesting this incident was part of a coordinated attack or state-sponsored action; however, Schober indicated that while it’s a possibility, it remains unverified. “The transportation systems, airlines, back-end reservation systems, and radar systems are, unfortunately, mostly quite outdated. They belong to obsolete systems,” Schober explained. Cybercriminals are well aware of this situation and will look for opportunities to exploit vulnerabilities. Whenever a disruption or even a minor glitch occurs, it presents them with an opportunity to strike.