Home

Tech Outages Are Only the Beginning. A.I. Will Bring More Disruptions.

Leave a comment

Click the link below the picture

.

Sometimes, especially lately, it can seem like computer software is built on a house of cards.

That is how the cybersecurity researcher Paulos Yibelo felt when he discovered a vulnerability in document and image processing software that is ubiquitous across the internet. The bug he found, called Upload2Shell, would enable a hacker to break into Google, Zoom and potentially thousands of other companies’ software just by uploading a file.

Mr. Yibelo, who recently founded a company called PWN AI, unearthed the bug by using artificial intelligence to meticulously probe what happens behind the scenes when people upload files to different services. He found that Ghostscript, software initially created 38 years ago to convert images to different formats, could be exploited by uploading files containing a custom font to execute malicious computer code. Mr. Yibelo reported the flaw to Artifex, the Ghostscript developer, and affected companies, which issued patches.

Mr. Yibelo’s finding, which has not been previously reported, is another example of how advanced A.I. has in a matter of months radically transformed cybersecurity and plunged the world into a disorienting new digital chapter in which severe technological disruptions may become far more commonplace than they were in the past, according to current and former U.S. and Western intelligence officials, security engineers and A.I. experts.

Those tremors have now bled into the daily life of Americans, as their email inboxes go dark and their flights are grounded. Viral rumors on social media have stoked confusion and fear about a widespread internet outage that could affect banking systems or critical infrastructure, even if recent events were not all A.I. related. It’s impossible for average people to tell what’s a malicious hack — could, for example, a 401(k) be drained? — and what’s simply antiquated software.

Over the past two weeks, a flood of outages slowed banking transactions, disrupted Microsoft products, interrupted Apple services, shut down a key system at the National Security Agency and temporarily grounded all commercial air traffic in the United States.

Officials have said they had no evidence that those episodes were caused by malicious cyberattacks. Instead, for a number of the outages, aside from the national air grounding, U.S. officials pointed to a more mundane explanation: Companies and government agencies are hardening their digital systems against new vulnerabilities, and these software patches are going awry.

In some of those episodes, including at the National Security Agency, the vulnerabilities at issue were detected with the help of A.I., according to U.S. officials. Individuals and businesses alike are using A.I. models to easily spot overlooked computer bugs and, in some cases, build proof-of-concept hacks in days or even hours that once might have taken scores of elite teams months to assemble.

At the same time, fears that bad actors are also using A.I. to work just as quickly have made defenders feel like they need to sprint to push fixes, which can result in messier updates that lead to disruptions.

Events last month were considered so alarming that some cybersecurity doomsayers (and opportunity-seeking marketing specialists) took to referring to the month as “Black October.” The label conveyed a sense that the world appeared to be suffering through a never-ending cascade of A.I.-powered software vulnerabilities that left defenders outmanned and exhausted.The gloom has been even worse this month, putting a public already anxious about technology’s rapid pace of acceleration further on edge.

It is unlikely to be a momentary blip.

Estimates vary, but cybersecurity experts have said that the next year or two could be especially perilous, as advanced A.I. models find coding bugs that humans never spotted before — even if many malicious hacks remain more mundane. Some hope that after that painful transition period the internet will reach an equilibrium as A.I. helps overwrite vulnerable computer code. Others hold the more jaundiced view that A.I. represents an interminable threat to the interconnected technology upon which modern life and the global economy operate.

The past week is “the new normal,” retired Gen. Paul M. Nakasone, the former head of the N.S.A. and U.S. Cyber Command, said at a conference held in Washington by Vanderbilt University’s Institute of National Security.

“Industries and agencies are trying to rapidly modernize and at the same time defend themselves,” Mr. Nakasone said. “A.I. is driving all of this — the opportunities and the challenges.”

Mr. Nakasone, who is an OpenAI board member, described A.I. as “the most disruptive technology” in history, which would have profound consequences not just for the digital world but also the rest of society. “We are in an era where we’ve got to think differently about how we’re going to be more resilient.”

Mark Zuckerberg, Meta’s chief executive, is credited with popularizing the Silicon Valley cri de coeur to “move fast and break things.” A more honest assessment of the internet might be that it was broken from the start, and moved too fast to ever be fixed.

The rise of the technology industry has been defined by fierce competition, as companies jockeyed to rush their products to consumers quickly while security, which can create friction and slow progress, often remained an afterthought. Deficiencies in hasty code build over time, a concept known as technical debt. The focus on rapid product development has left many tech products riddled with hidden security holes that, at least historically, only a few determined hackers might find and exploit.

“The amount of technical debt that most companies will allow and live with is unforgivable,” said Juan Andrés Guerrero-Saade, a distinguished fellow for security and A.I. research at the cybersecurity firm SentinelOne. “Folks have been riding until the wheels fall off when it comes to technology. There’s always money in releasing, and there’s no money in maintaining.”

A.I. has upended the playing field, making it much easier to find the cracks in popular software. New cyber-focused A.I. models that can ferret out vulnerabilities have whipped some bug hunters into a frenzy as they compete to find novel ways to break into systems. Companies have also bought in, using A.I. to blast their own infrastructure and find bugs before hackers can.

“You’re essentially taking these brittle code bases and shaking them,” Mr. Guerrero-Saade said.

Even the N.S.A., which specializes in hacking and cyberdefense, is not immune. It had a days-long classified network outage this week because it installed faulty updates that sought to address flaws in the PHP programming language that were uncovered by A.I.-wielding researchers, officials said.

This week, Senator Maria Cantwell of Washington, the senior Democrat on the commerce committee, and Senator Mike Rounds of South Dakota, the Republican chairman of a cybersecurity subcommittee, wrote a letter to the White House warning that A.I.-enabled cyber threats were outmaneuvering America’s defense systems.

“A.I. is rapidly increasing the ability to discover and exploit weaknesses in software, and our defenses are not keeping pace,” the senators wrote in a letter to Jay Clayton, the national intelligence director, who also leads the White House super intelligence task force. “We cannot meet these threats with yesterday’s cybersecurity infrastructure or rely on individual companies to solve a problem that spans the entire digital economy.”

Recent data shows just how dramatically A.I. is transforming the cybersecurity ecosystem. Microsoft released over 1,400 patches for its various products last month, a rate about 10 times higher than its historic baseline. The tally eclipsed the previous record of nearly 1,000 set just a month earlier, in September.

Experts worry that there are fewer qualified humans available to triage the bug explosion. Major tech companies have made mass layoffs this year, in part because they have sought to replace workers with A.I., leaving them without institutional knowledge about how their old code was written.

New hiring data released this week by LinkedIn and reviewed by The New York Times showed that despite the rise in cybersecurity incidents and more bugs spotted than ever, U.S. cybersecurity hiring has underperformed relative to overall hiring.

Lea Kissner, the chief information security officer at LinkedIn, said that the current cybersecurity challenges meant companies needed humans with expertise more than ever.

“No one knows the best way to do security right now, because everything is changing so quickly,” said Dr. Kissner. “So you desperately need people who understand how things work, and have curiosity, and know how to solve problems.”

Despite the deluge of bugs exposed by A.I., some things haven’t changed. Most recent major cyberattacks have demonstrated hackers don’t need A.I., or only use it on the margins, to achieve their goals.

Iran’s targeting of scores of municipal water systems in the United States this summer largely relied on rudimentary tactics that seized on weak or default passwords on the computers that control operations. A criminal group’s hack of the F.B.I.’s sensitive personnel files relied on unpatched Oracle software that had been available for months. And recent attacks on refrigeration systems in Europe also appear to be relatively unsophisticated.

.

https://static01.nyt.com/images/2026/11/21/multimedia/21DC-OUTAGE-topart-sub-wghl/21DC-OUTAGE-topart-sub-wghl-jumbo.jpg?quality=75&auto=avifJPMorgan Chase, the nation’s largest bank, reported trouble processing transactions earlier this week, one of several tech outages with unclear causes that have rattled the United States in the past several weeks. Credit…An Rong Xu for The New York Times

.

.

Click the link below for the complete article:

https://www.nytimes.com/2026/11/21/us/politics/tech-outages-ai-disruptions.html

.

__________________________________________

Leave a Reply