Summary
David Ondrej reviews what he calls Black October: a 48-hour stretch of severe vulnerabilities and hacks that he attributes to swarms of AI agents scanning the internet for flaws. He points to a chart of reported CVEs that has gone nearly vertical in recent weeks, and says about 30 severe issues surfaced in a single day across PHP, Node.js, Nginx, Go, GitLab and Moodle.
He then unpacks the three worst incidents. Two 10-out-of-10 PHP remote code execution flaws affect software behind WordPress, MediaWiki, Magento and Laravel, and details reportedly leaked about 15 hours before a patch; Cloudflare saw attempted exploits climb into the millions. A 12-year-old GitLab vulnerability also came to light. He also covers eMudhra, a certificate issuer whose private keys were generated from tiny numbers, including the number four, allowing anyone to forge certificates.
The second half is practical. Ondrej advises running the best frontier models from Anthropic and OpenAI at maximum reasoning effort against any code before shipping, and shows the always-on agent loops he runs against his own repositories, including a supply chain security sweep that monitors news, GitHub and social media. His central argument is that defending against agent-driven attacks requires running your own agents against your own code.
📺 Source: David Ondrej · Published October 23, 2026
🏷️ Format: News Analysis
