AI found the bugs. Rushed fixes are tied to NSA and bank outages
Bank of America, Wells Fargo and JPMorgan all stumbled this week, and Senator Mark Warner's office ties botched AI-driven patching to an NSA comms outage too.

Advanced AI models are surfacing software vulnerabilities faster than IT teams can safely patch them, and the botched rollout of those fixes has been linked to an outage of the NSA's internal communications system and this week's outages at Bank of America, Wells Fargo and JPMorgan Chase, according to a spokesperson for Senator Mark Warner's team. For executives, defensive AI has become a change-management and governance problem rather than a procurement line item, with new legislation from Representative Andy Ogle already in motion.
Start with what broke. Over the past week, flights were grounded, emails stopped arriving, and bank transactions failed across the country, and neither politicians nor businesses offered a clean explanation. Officials did rule out one fear: this was not a coordinated cyberattack. But AI still had a hand in it, just not the hand anyone expected. Advanced AI models are now discovering software vulnerabilities at a speed humans cannot match, and tech teams are racing to patch those holes before bad actors use equally powerful AI tools to exploit them. The botched implementation of those patches has been linked to an outage earlier this week of an internal communications system at the National Security Agency, as well as the banking outages, a spokesperson for Senator Mark Warner's team told Fortune. Bank of America, Wells Fargo, and JPMorgan Chase all suffered outages this week. The Washington Sun also reported that bank CEOs and the NSA told Congress that AI-driven patches were to blame.
The irony is sharp enough to sting: the tools built to prevent outages appear to be causing them, or at least manufacturing a new class of human error. "It feels like every day brings another reminder that AI is moving faster than our ability to manage its consequences," Warner said in a statement. That is the uncomfortable shape of this story. Defensive AI is not the villain. The way companies are deploying it is.
The scale of the scramble shows up in the hiring. Bank of America disclosed last month that it had brought on 1,000 contractors to patch flaws exposed by Anthropic's Mythos model, on top of 3,000 internal staffers, CNBC reported. That is 4,000 people chasing bugs a model surfaced in seconds. "This is a failure of implementation," Amit Shah, founder and CEO of InstaLILY AI, tells Fortune. "The AI finds a vulnerability in seconds and may not know what else breaks when you fix it. The person approving the patch sees a recommendation, not the systems it touches. So a tool built to prevent outages can end up causing them."
Jeff Collins, CEO of the IT technology platform WanAware, frames the driver as fear. Anxiety around "the theoretical risk of a breach" is prompting a "rush to eliminate" vulnerabilities, he tells Fortune, and the mismatch is structural: "AI now discovers vulnerabilities at machine speed while our testing and deployment safeguards remain stuck at human speed." Palo Alto Networks CEO Nikesh Arora said on X that IT teams are drowning in fixes. "Some of the recent outages are a consequence of a deluge of patches," he wrote, adding that between the mandate to plug holes and the threat of AI-assisted hacks, "we have a challenge," with "no silver bullets."
Who owns the mistake is genuinely murky, and that ambiguity is its own risk. A human may have approved a fix without vetting its blast radius, but it is unclear what the AI tool told them. Did the model overstate a vulnerability's severity and create panic? Did it recommend changes en masse that a human waved through? Are security managers pushing AI adoption faster than their teams can absorb it, to avoid a potentially catastrophic breach? Collins argues the framing is wrong: "The failure isn't that AI patches too aggressively or that exhausted engineers rubber-stamp fixes, but that we plugged a hyper fast diagnostic engine into legacy operational pipelines never built to absorb that velocity of change."
The vendors are moving deeper in, not backing off. OpenAI sells cybersecurity-specific models to help organizations find vulnerabilities at scale and is rapidly enrolling customers into its Daybreak program, which gives approved security professionals access to its most powerful defensive cyber tools. The initiative's stated goal is to "patch the planet," and its products include dashboards that classify bugs by severity to help teams triage them. Anthropic's version of Daybreak is called Project Glasswing. On Nov. 5, a week and a half before the JPMorgan outage, OpenAI president and co-founder Greg Brockman announced a partnership with JPMorgan Chase to use OpenAI models to help secure its systems. An OpenAI spokesperson said the JPMorgan outage stemmed from legacy systems getting overwhelmed by code modernization work. A JPMorgan Chase spokesperson said OpenAI's tools were not a factor, tracing the problem to "a temporary issue related to a software bug in on-premise hardware systems," and added that the company did not identify that bug through AI tools or fix it with AI.
Brockman has told The Information there was "no question" that cyber would eventually become a bigger business for OpenAI than coding. The company swapped out its chief revenue officer earlier this year for Dali Rajic, tasking him with using his cybersecurity enterprise sales experience to accelerate that line. Read that alongside the outages and the incentive picture sharpens: the vendors selling the fix are also scaling a business built on the problem staying large.
Lawmakers are already moving. Representative Andy Ogle plans to introduce legislation today calling for better coordination as AI speeds up the discovery and exploitation of software flaws, NextGov reports. For executives, the lesson is not to abandon AI-powered defensive security. It is that patch velocity has become an operational risk of its own, and the controls around it, testing, staged rollouts, clear ownership of approvals, have not kept pace. Boards that treat AI security tooling as a procurement decision rather than a change-management problem are the ones most exposed. The next outage probably will not come from a hacker. It will come from a fix that shipped too fast.
This story's Key Insights and Take-aways are locked.
Create a free account to unlock Executive Actions for one credit.
Register to UnlockAlways free for Executives Club members. Join the Club
More in Business
S&P 500 Enters Correction, Down 10% From Its High on AI Worries
The index's slide past the 10 percent line is a symbolic verdict on investor mood, and the AI trade that powered the rally is now the thing testing it.
Dow and S&P 500 enter correction as Nasdaq logs worst week since 2020
All 11 S&P 500 sectors fell more than 1% in the same week for the first time since April 2025, a breadth signal that matters more than any single index level.
Chip stocks lose $2.3 trillion in a week, worst run in 25 years
A $2.3 trillion weekly wipeout in semiconductors is forcing every board with an AI story to ask whether it owns a thesis or just a crowded trade.



