AI is moving into your browser. Who controls it?

AI Browsers are the next great consumer software battle. The terms of the deal are being written right now. Produced in partnership with Shift Browser

J Studios / Getty Images

By now, anyone using AI seriously has gotten used to browser bloat. A window for the doc, another for Claude, another for ChatGPT, another for Gemini… Oh, and a few more for hands-on research, plus a final one for whichever tool briefly seemed like the answer last Tuesday.

We fought to be able to work from home, and yet we find ourselves saddled with a new daily commute: jumping between our tabs. Copy, paste, re-prompt, re-explain, re-upload. Our AI tools live inside walled gardens, reducing us to glorified messengers. It’s frustrating. Instead of creating windows for AI to see through, the industry decided to start giving it a sledgehammer.

Earlier this week, the founder of a SaaS company called PocketOS sat down at his computer, only to discover his entire production database was gone. So were the backups. The whole thing had been deleted in a mere nine seconds, by an AI agent that wasn’t authorized to do anything of the sort.

The agent was Cursor, running on Claude Opus 4.6. It hit a credential mismatch and independently decided the cleanest fix was to delete a Railway volume. The agent went searching for an API token, found one in a file that had nothing to do with the task at hand, then used it to issue a single destructive API call. The command went through without an environment check or confirmation prompt. Not even an “are you sure.” It took production and the backups down together.

When the........

© Quartz