Here's a theory for why everything is breaking down

Anthropic PBC made a record-breaking debut on Nasdaq on Thursday. Credit: AP/Andres Kudacki
This column reflects the personal views of the author and does not necessarily reflect the opinion of the editorial board or Bloomberg LP and its owners. Dave Lee is Bloomberg Opinion’s U.S. technology columnist. He was previously a correspondent for the Financial Times and BBC News.
It was perhaps inevitable that Anthropic PBC’s record-breaking stock market debut would collide with the sense of a cyber meltdown over the past week. The technological leaps in artificial intelligence are the backdrop for both Anthropic’s $1.77 trillion valuation — which rests on the assumption that AI’s impacts are only just beginning — and the current state of digital emergency, in which planes are grounded, spies can’t communicate with one another, Microsoft Office seizes up and bank apps stop working.
To all this, President Donald Trump says: "Relax. Everything’s good."
Setting aside his blitheness and the communication circus of his administration, Trump is right that the recent outages are not the result of a coordinated cyberattack but appear to be separate instances of software (or hardware) gone awry. But it’s hard to pass on that message of reassurance to friends and family who have been asking variants of the same question over the past few days: "Why is everything breaking?"
Is it really just a coincidence, many reflexively wonder, that as companies cut employees and turn to AI, technology is failing for unrelated reasons? "Americans aren’t stupid," wrote Tricia McLaughlin, former Homeland Security spokesperson. "We want and deserve to know what’s really going on."
Fine. Here’s what’s really going on. The presence of AI, of models that can seek out bugs no human has been able to find, has changed everything. It’s finding bugs that have been hiding in software, sometimes for decades, in widely used programming languages like PHP, which power most websites. It’s a perilous situation with real risks. Hedge fund Bridgewater said that cyber warfare was on an "exponential path" and that it was managing money "based on a one-in-three chance of an economically destabilizing attack within the next year."
Data from the CVE Program, which logs confirmed vulnerabilities in software, shows 2026 has been a banner year for bug discovery. More than 100,000 have been logged, with more than half graded as either severe or critical. With more than a month to go, 2026’s bug count is more than that of 2025 and 2024 combined, according to an online tracker.
You can count on 2027 eclipsing this year’s total, said Joe Levy, chief executive officer of cybersecurity firm Sophos. "But the silver lining here," he said on a Zoom call, "is that what I think we’re seeing with all of these disruptions is a response from the industry to go ahead and get that patched as quickly as possible."
That carries its own problems. Suddenly jumping into an antiquated system to issue a fix is like hurling yourself at the heaviest weights after a six-month break from going to the gym, Levy said. The bad news for the U.S. government is that it may be particularly vulnerable with decades-old systems like COBOL sitting beneath important functions like collecting taxes or paying Social Security. The $50 billion request to modernize air traffic control is just the beginning.
AI companies are being aggressive in seeking out these vulnerabilities. Anthropic recently said it was shortening its window for making discoveries public. It and other AI labs are offering free assistance to fix discovered bugs. Still, you might accuse these firms of being arsonists returning with a bucket, seeing AI the same way Homer Simpson famously described alcohol: the cause of, and solution to, all of life’s problems.
Whether the criticism is fair is irrelevant now that the cat’s out of the bag. Anyone with a capable-enough AI model, including those that are open weight and unable to be reined in, can and will find these flaws. Last week, the FBI warned that hackers were increasingly using a system called ARTEX to set off autonomous penetration, targeting financial services, government, tech, defense and several other industries. It works by harnessing several AI models, including those made in China. The FBI warned intrusion efforts had progressed from "initial targeting to privileged access in under three hours."
In other words, a hacker — with minimal skill — can just send an AI out to find any vulnerability wherever it can by relentlessly hitting vulnerable systems and seeing what lands. That could be anything from websites to transportation systems to hospitals. In the past, the time and effort needed to discover holes and exploit many critical systems simply wasn’t worth the payoff, but that’s no longer the case. "Obscurity used to buy you years, maybe decades," said cybersecurity researcher Brendan Dolan-Gavitt. "Now it buys you days." (Dolan-Gavitt was giving a presentation at a cybersecurity conference, during which he used AI to reverse engineer the code that prevents Photoshop users from forging currency.)
The only way to fight AI-enabled hacks is with AI-enabled defense — and the money required to implement it. Americans should be prepared for the disruptive consequences. "The patch needs to go instantly from being available to being patched in production, because it’s being exploited in the wild," Microsoft Corp. CEO Satya Nadella said at the New York Times DealBook summit this week after apologizing for his company’s outages. "That’s going to create instability."
Further, Nadella warned, some infrastructure will be considered "unpatchable" and will presumably need to be replaced at great cost, a national mobilization to set right the crumbling bridges and leaking dams of the digital realm. Maintainers of open-source software, often individuals or nonprofits, must be given support so these critical building blocks of modern software don’t become a weak link or a bottleneck to better security broadly.
Where possible, federal agencies should seek to coordinate updates to minimize disruption, like a city planning road closings. But things will inevitably go wrong. When they do, plain and simple explanations — what we might simply call a heads-up — would do much to ease fear and stop the financial markets from overreacting, hopefully going some distance to make those who "just ask questions" decide to hold their tongues for once.
And we should all try to rest easier at night by reminding ourselves that what’s going on is as much a sign of things being fixed as anything else. "In two, three years’ time, I think America’s cybersecurity will be much stronger than it is today," said Google’s president of global affairs, Kent Walker, when I met with him this week. The foreseeable future will bring frustrations and inconvenience, but a digital emergency need not be a cause for panic — just action.
This column reflects the personal views of the author and does not necessarily reflect the opinion of the editorial board or Bloomberg LP and its owners. Dave Lee is Bloomberg Opinion’s U.S. technology columnist. He was previously a correspondent for the Financial Times and BBC News.