7 min read
Did OpenAI’s Agents Really Go Rogue on the Australian Government?
"Rogue" is a great word for a headline and a terrible word for an incident report. WIRED used it to describe the…
Artificial Intelligence
4 min read
Your Personal AI’s Memory Now Lives on Google’s Servers—Inside a Sealed Box
Your AI assistant can now remember context between sessions, but that state lives in Google's cloud, protected by confidential computing promises instead…
6 min read
Why Did Your OpenAI Bill Jump to $40,000 Overnight?
When your AI billing dashboard jumps from $200 a day to $38,000 overnight, you haven't been hacked—someone just found your API key…
8 min read
Should You Hand an AI Agent Your Credit Card and Your Inbox?
The $64 mistake isn't the scary part—it's proof that an autonomous agent will spend your money on imperfect judgment, unsupervised, and then…
4 min read
When ChatGPT Voice Sends Your Email, What Did You Actually Agree To?
The three GPT-6 model names in the headlines are the van. The OAuth scope is the key you handed the courier—and you…
7 min read
Three things broke when I enabled GPT-6 caching—the fourth cut my bill 70%
GPT-6's new caching API lets you mark exactly where stable context ends and volatile input begins—but only if your prefix is actually…
6 min read
What Happens When the AI Agent Holding Your Card Takes Orders From a Stranger?
Ars Technica reports that Meta's Muse assistant—launched with Shopify integration and payment authority—has an unpatched vulnerability that combines social engineering with apparent…
4 min read
“The Model Said So” Is Dead. Provenance Won.
An AI hallucinated nuclear weapons on a Chinese vessel, nearly causing a military boarding. The fix isn't a smarter model—it's architecture that…
6 min read
The Demo Era for Foundry Agents Is Over. The Kill-Switch Era Just Began
Business units have been spinning up AI agents with nobody able to say no. Microsoft just shipped the first governance control—a tenant-wide…
6 min read
Perplexity Told Its Engineers to Check In on the Model Less Often
The alarming line in OpenAI's Perplexity case study isn't that the AI changes software—it's that engineers now "check in less frequently" on…