Vibe coding will bankrupt businesses | David Gerard

unconvincingDavid Gerard (The Tech Report)podcast2026-10-26read planted ai llmssecuritysoftware engineering

Synopsis — AI-drafted from Dan's notes

Isaac Pound talks to David Gerard of Pivot to AI about the previous week’s PHP scare. The bugs were real: two critical remote-code-execution flaws in PHP’s memory handling, rooted in a problem a core developer had first spotted in 2024 and that stalled for months. The coordinated release was set for 22 October and pulled forward a day after leaks: the CVE numbers turned up online, the security firm Calif accidentally posted a video of its exploits, and Chainguard shipped patched images more than six hours early. Gerard, a former sysadmin who patched his own WordPress sites that night, calls it a serious bug wrapped in an overblown panic. Big sites would always update; the exposure is neglected small sites and appliances with PHP interfaces that never get patched. He says he heard of no real exploitation, though attempts were in fact seen in the wild (Amazon’s honeypots from 17 October; CISA listed both flaws as exploited on 26 October).

He blames the panic on AI marketing: claims that AI would turn the patch into an exploit in minutes, floods of chatbot-written “exploits” that did not work, and labs implying credit for finds they had no part in (his example is an OpenAI post about a Brotli bug). Worth knowing: by the PHP Foundation’s account, Calif and OpenAI did both report this PHP bug independently on 11 August. The week’s other story was worse: eight eMudhra intermediate certificate keys were trivially weak, one of them literally the number 4. Browsers distrusted them and eMudhra is leaving the TLS business. He suspects a chatbot’s idea of a random number but says nobody knows that AI was involved.

Isaac recalls Bruce Schneier predicting that AI makes security worse before it gets better. Gerard agrees, because enterprise security is so poor that cheap finds are everywhere, and he calls chatbot scanning expensive output that looks like a fuzzer’s. Then the title. There is money to write code and none to maintain it, and vibe coding piles up technical debt no one can read. When the bubble pops and tokens get dear, local models will be too slow to help, and he expects businesses to go broke when software they cannot maintain stops working.

Connections