AivexaNewsSearch
AI news for builders and product teamsChecked every hour

Will AI expose the rickety infrastructure of the internet?

Collected Oct 26, 2026

TechCrunch's Equity podcast devoted a segment this week to a sweeping security scare tied to the open source language PHP, after an alert told site owners to patch immediately or, if they could not update in time, take their sites offline. Hosts Kirsten Korosec and Sean O'Kane were joined by weekend editor Anthony Ha for the conversation, which ranged from the technical scale of the problem to what the episode suggests about the state of the web's foundations.

PHP, formally Hypertext Preprocessor, underpins the content management systems behind a large share of the web, including WordPress and Drupal. Korosec put the affected footprint at roughly 70% of websites, or tens of millions of sites. That install base is the core of the story: a flaw in a language this widely deployed is not a single-product bug but a problem distributed across hosting providers, agencies, small businesses and government pages, many of whose owners may not know PHP is running underneath their site at all. Ha described the underlying vulnerability as one that had existed in PHP for decades without being publicly known.

O'Kane framed the moment as a reckoning over the downside risk of building so much of the internet on open source technology. His image for the situation was that at the bone level the internet is held together with glue and dreams, and he noted how often the last line of defense has been an individual maintaining a tiny repository as a hobby who happens to notice something is wrong. The new variable, in his view, is scale: widely available large language models and agents that can be pointed single-mindedly at a task make vulnerability hunting far cheaper.

A central tension the panel explored is disclosure. According to the discussion, researchers participating in an OpenAI and Wiz initiative, with Wiz now owned by Google, found the flaws, and Ha issued a correction on air to clarify that the discoveries came from human researchers who used AI to exploit the bug rather than from models working alone. That correction matters for how the industry reads the episode: it is evidence that AI-assisted security work can surface deep flaws, not proof that autonomous agents did so here. The logic driving the rushed announcement, as the panel described it, is that if AI-assisted tooling can find a decades-old hole, others can too, and the only route to a patch is public disclosure. Sean also noted criticism of how the rollout was handled, including what some saw as self-congratulatory framing in OpenAI's post, and said a couple of people got ahead of the embargo, setting off a race among attackers.

Signs of exploitation came quickly. Cloudflare CEO Matthew Prince posted that attempted hacks of the PHP flaw had escalated to the millions, according to O'Kane. Korosec pointed to an Amazon blog post describing a honeypot, or decoy, network that recorded attempts to exploit the vulnerability before it was publicly announced. She drew the contrast that stood out to her: Amazon has the capital, tooling and staff to detect and respond, while many sites, potentially including government ones, have nothing comparable. An unpatched WordPress install on shared hosting is, in practice, the kind of target that has neither a security team nor a honeypot to warn it.

Korosec raised a second-order cost that is easy to miss: link rot. Owners who cannot patch may simply shut sites down, so pages that carried a piece of institutional or neighborhood history could vanish rather than be fixed. That is a durable loss even after the technical patch is applied elsewhere, and a reminder that remediation advice lands unevenly across operators with different resources.

The panel also debated whether this is the start of a pattern of uncovering other aging software. Korosec said she expects new startups to emerge around this kind of security layer, and pointed to open source tooling from companies such as Nvidia as one avenue for adding protection. She also raised concerns about how far the risk could reach, including banks and privacy implications. Those are forward-looking worries rather than reported findings, and the episode did not present evidence of exploitation at that scale.

Why it matters: developers and product teams should treat this as a prompt to inventory where PHP sits in their stack, including inherited and third-party sites, and to confirm patching paths exist before the next alert rather than after it. Inference on my part: the durable shift is not this one bug but the economics of discovery, since AI-assisted research lowers the cost of finding old flaws and raises the premium on knowing quickly which of your systems are exposed. Smaller operators and agencies without dedicated security staff are likely to feel the squeeze first.

Read at TechCrunch · AI

Based on reporting from the original publisher. Visit the source for full context and later updates.

Publisher excerpt

The moment you realize the internet is "held together by glue and dreams."