i
News
News · 2026-11-18

Banks and NSA say AI may have caused major outages

Dataist Dataist

In November, major outages took Microsoft and Apple services offline, disrupted US banks and caused delays and cancellations on United Airlines flights. All air travel in the country stopped for several hours, and this week the US National Security Agency’s (NSA) internal network for classified information went down. Washington Sun sources say representatives of the banking sector and the NSA told Congress the outages may have been caused by updates to systems designed to protect against AI-powered attacks.

Cover: Banks and NSA say AI may have caused major outages

Updates overloaded systems

Lawmakers are trying to determine what caused the major outages and the growing number of large-scale breaches. Early information suggests suspicions of AI involvement may be justified, though preliminary accounts point to something other than a direct attack.

Technology companies and federal agencies are strengthening their defenses, using AI to find vulnerabilities that attackers could exploit. But organizations whose systems underpin critical services may have overloaded them by trying to deploy too many updates at once.

A Washington Sun source in Congress said models are finding numerous vulnerabilities, prompting efforts to release fixes as quickly as possible. But fixes are piling up faster than systems are designed to handle them.

US Homeland Security Secretary Markwayne Mullin also said on Tuesday that new technologies are being added to outdated computer systems, overloading them. He estimated there was a 98% chance that a cyberattack was not behind the incident.

AI is accelerating the vulnerability race

That explanation is consistent with officials’ assurances that the outages were not related to cyber threats. But the rush to install updates leaves open the question of whether hackers are preparing new attacks.

Experts have long warned that AI has a dual effect on cybersecurity. It helps researchers find weaknesses, but attackers can use the same tools. Earlier this year, Anthropic said in a report that its Claude Mythos model had found more than 10,000 vulnerabilities in existing software. Companies are rushing to fix such flaws, and that race could pave the way for new attacks.

Nathan Calvin, director of strategy at Encode, said the absence of an attack could be seen as encouraging. But the current events show that critical infrastructure depends on software riddled with vulnerabilities and running on outdated systems. Calvin believes the race will continue as models advance.

Dataist's take: The outages show that using AI to find vulnerabilities does not, by itself, protect infrastructure. Fixes still have to be deployed safely in systems, some of which are outdated. As models and update releases get faster, the mismatch between new technology and old infrastructure becomes more visible.

Daily AI news

Every day we pick what actually matters in AI and explain it plainly — no hype, no filler. Subscribe if you want to follow where the industry is going.

Only what matters — every day

Follow on X