AI Readout

An AI run found 948 ways to take over a Linux machine. Patch yours

OpenAI's cyber model wrote 948 Linux root exploits in one run, Google's video model now makes 15-second clips in Vids, and Meta paused its new business terms.

One robots.txt line will tell four AI labs not to train on your site

Four AI labs agreed on one way to say no training, a tampered Claude Code extension reached Cursor users, and GPT-6.1 Terra costs half what the last Terra did.

Patch your website's PHP this week, and Anthropic opens its books

Two PHP holes reach WordPress before login, Pennsylvania wants labels on misleading AI ads, Anthropic's S-1 shows $16.3 billion in half-year revenue, and my weekly check failed for two months.

Meta's new business terms reach your website on October 30

Meta's crawling terms cover anyone with a Page, cheap AI agents ran 1,100 break-in attempts, Gemini 4 Argon's price doubles December 1, and my new tagline was someone else's.

Anthropic cut its test agents off from the internet, and more code isn't more software

Anthropic's models filled in real web forms during tests, a Harvard study finds AI code piles up at review, OpenAI posts 719 math papers, and my model picker had a dead tier.

ChatGPT stopped answering in paragraphs, and Wikipedia found OpenAI's agents