---
title: "AI Found Two Critical Flaws in Most of the Web, and Security Execs Are Spooked"
description: "Anthropic’s security chief warns of a coming vulnerability tsunami as AI turns obscure bugs into easy exploits"
author: "Nate Ledger"
published: 2026-10-23T10:29:53Z
modified: 2026-10-23T11:40:30Z
url: https://rews.cc/a/ai-found-two-critical-flaws-in-most-of-the-web-and-security--2416cc
language: en
tags: ["ai", "cybersecurity", "php", "vulnerability", "infrastructure", "tech"]
publisher: "Rews (https://rews.cc)"
---

# AI Found Two Critical Flaws in Most of the Web, and Security Execs Are Spooked

*Anthropic’s security chief warns of a coming vulnerability tsunami as AI turns obscure bugs into easy exploits*

By Nate Ledger · October 23, 2026 · https://rews.cc/a/ai-found-two-critical-flaws-in-most-of-the-web-and-security--2416cc

## In brief

- AI helped find two critical flaws in PHP, the language up to 70% of websites run on, sending companies scrambling
- Models can find old vulnerabilities at machine speed and hand exploits to people who couldn't have written them
- Box CEO Aaron Levie called it the death of "security through obscurity" for legacy systems
- Anthropic's Jason Clinton warned of a coming "vulnerability tsunami" and said critical infrastructure hasn't caught up
- Anthropic's deputy security chief appears to see roughly 18 months of structural industry exposure

It has been, to use the technical term, a rough week for the people whose job is keeping the internet from falling over. AI helped discover two “critical” flaws in the infrastructure of PHP — a programming language that up to 70% of websites run on — and [Business Insider reports](https://www.businessinsider.com/what-smart-people-say-cybersecurity-openai-internet-threats-2026-10) the discovery sent companies scrambling to patch before someone less friendly found the same holes. Which, given how the discovery happened, they very well might.

Here is the mechanism, and it’s worth understanding because it explains why everyone from Anthropic’s security chief to Marc Benioff spent the week posting in increasingly alarmed fonts. Software is full of old vulnerabilities. It has been full of old vulnerabilities for decades. This was, in a strange way, fine — exploiting those vulnerabilities took skill and time, both of which were scarce among attackers, and patching them took money, which was scarce among everyone else. So the bugs sat there, a kind of mutually accepted truce between the people who could theoretically break things and the people who couldn’t afford to fix things.

![A magnifying glass reveals red damaged areas on a dark blue padlock with holes and stitched patches against a teal background.](https://rews.cc/img/4420c20350b88bcbd761d0687de615db91d1f8e5.webp)

AI breaks that truce at both ends. Models can find vulnerabilities in old code at machine speed, and they can hand the exploit to someone who couldn’t have written it themselves. Deedy Das, a partner at Menlo Ventures, declared on X that “the era of vibe hacking is here” — he said he’d seen firsthand that “even non-technical people” can use AI models to exploit common vulnerabilities like PHP’s, essentially pointing a model at a website and telling it to find the holes. “Once you figure it out, it is downright trivial to do,” he wrote. “The digital infrastructure of the world is at risk.”

Aaron Levie, the CEO of Box, framed it as the death of “security through obscurity” — the lazy but historically effective strategy of hoping no one with the requisite skills ever looks too hard at your legacy systems. AI agents, he wrote on X, can devote once-unthinkable time and effort to looking at exactly that. His to-do list for companies: upgrade legacy infrastructure, modernize security policies, tighten access controls. Though Levie also found the cloud inside the lining, predicting a boom in cybersecurity jobs as companies hire humans to point the agents at critical work, review their findings, patch systems and respond to incidents. “Agents will help on the defensive side,” he wrote, “because they can both accelerate this work and equally simulate the risk vectors.” It is not every CEO who responds to a threat by spotting the headcount opportunity, which is a very CEO response.

![Two small figures use oversized needles and red yarn to mend a cream knitted sweater, while a large robotic hand pulls threads from its detached sleeve.](https://rews.cc/img/485a32f960243dd51210214520b836a089592660.webp)

The bluntest warnings came from Jason Clinton, Anthropic’s deputy chief information security officer, who said a “vulnerability tsunami” is coming and that critical infrastructure hasn’t caught up. After posting Wednesday on X about the need to “patch critical infrastructure,” he went further Thursday: “As bad as things have gotten in the past 4 days, they are going to get a lot worse before they get better.” His advice to security leaders: “never let a crisis go to waste: please go to your boards and ask for help right now. not for ai, for the basics. (we need to hold on for ~18 months.)”

Translate that parenthetical, because it’s the interesting number in the whole package. Roughly eighteen months is how long Anthropic’s own deputy security chief appears to think the industry is structurally exposed — long enough, presumably, for defensive AI and patch pipelines to catch up, but not before the offensive tools get a long turn at bat. Marc Benioff, the Salesforce CEO, endorsed the alarm in his trademark laconic style: “Jason is 100% right. Code red,” plus a red alarm emoji.

Meanwhile the threat is not theoretical. Andrew Curran, a prominent AI analyst, wrote Wednesday — responding to an NSA cybersecurity advisory — that “regarding potential attacks on energy, food, and water infrastructure, the threat is quite real. Someone is currently probing infra and feeling it out, it is an active threat.” He also highlighted a hack of Quest Diagnostics, which he said serves “roughly half” of the US’s hospitals and physicians. Energy, water, food, hospitals: the list of targets is essentially the list of things you’d prefer not to lose at the same time.

![A factory complex sits atop a four-socket power strip, with three cables connected and smoke rising from a striped chimney against a blue background.](https://rews.cc/img/11c76719c36eb239c07ab19d58a88968cb232f85.webp)

Matt Shumer, the former Hyperwrite CEO whose February essay “Something Big Is Happening” went viral, returned Thursday with a new one: “The Internet Is Under Attack. Here’s How to Protect Yourself and Your Family.” “Right now, as you read this, swarms of AI agents are likely trying to break into the websites you use. Not one of them. Many of them,” he wrote, calling it “the biggest cybersecurity moment in years.” He sketched the personal version of the tsunami — leaked data making scam calls sound real, accounts hijacked through a reused password on some forgotten site, apps and pharmacies going down for days — while conceding the apocalypse hasn’t arrived: “the worst-case scenarios people are most scared of haven’t happened. I hope they never do.” His advice was decidedly pre-digital in flavor: update everything, treat every link and code as suspect, and “close the doors you aren’t using” by unplugging smart gadgets and deleting unused apps. “Every connected thing is another way in.”

The more measured voice belonged to Sriram Krishnan, a venture capitalist and former senior AI policy adviser to President Donald Trump, who called the PHP flaws a “harbinger of things to come” as models keep surfacing long-buried bugs. The software supply chain, he wrote on X, “will need the muscles to rapidly build and deploy fixes,” possibly through new public-private partnerships. Which is reasonable, and also an odd thing to need to build in a hurry — an industry suddenly discovering it requires emergency plumbing on the whole internet at once.