The AI Daily Grindhttps://aidailygrind.com/The AI news that matters, freshly ground every morning. In simple words.enhttps://aidailygrind.com/static/logo-512.pngThe AI Daily Grindhttps://aidailygrind.com/AI finds the holes, AI picks the locks, and the agents want your wallethttps://aidailygrind.com/2026-10-22.htmlhttps://aidailygrind.com/2026-10-22.htmlThu, 22 Oct 2026 10:30:00 +0000François GravelGood morning, and welcome to The AI Daily Grind. Today feels like a security day. AI tools from OpenAI and Google's Wiz found two very old, very serious holes in PHP, the language behind a huge part of the web. Then a leak gave attackers a head start. On the same day, Microsoft said a China-aligned…Good morning, and welcome to The AI Daily Grind. Today feels like a security day. AI tools from OpenAI and Google's Wiz found two very old, very serious holes in PHP, the language behind a huge part of the web. Then a leak gave attackers a head start. On the same day, Microsoft said a China-aligned group used AI agents to break into thousands of companies through ConnectWise.

This is the big pattern. The same AI that helps defenders also helps attackers, and they move at machine speed. Meanwhile, Anthropic launches Claude Duet, its new bet on agents, while a Transformer investigation shows Claude can still explain how to manipulate an election. Stores now fight over whether AI agents may shop for you. And the big labs spent record money on lobbying in Washington. More power for machines, more power for a few companies, and safety that is running to keep up.

There is good news too. Open models keep getting smaller and better, and normal people use AI coding tools to clean up their own TVs. Bon, watch two things this week: how fast websites patch PHP, and who writes the rules for AI agents that spend money and click buttons for us.

]]>