Snippets.

// AI Briefing

October 3, 2026

AI Briefing

OpenAI wants your next coworker to be an agent that never logs off, and this week's other stories show why that's both exciting and nerve-wracking: coding agents quietly dumped thousands of private screenshots onto public GitHub, and Apple is rewriting Mac permissions with agents in mind. Meanwhile TSMC's Texas plans show the physical cost of all this ambition.

OpenAI's New 'Dots' Are ChatGPT Agents That Keep Working After You Close the Tab
01ProductBetaNews ↗

OpenAI's New 'Dots' Are ChatGPT Agents That Keep Working After You Close the Tab

At DevDay, OpenAI launched dots: always-on agents powered by GPT-6 Astra, each with its own cloud computer and browser. You hand one a goal, connect apps and set what it may do alone, and it returns results for review, from fixing bugs and opening pull requests to updating sales proposals. It matters because it shifts ChatGPT from a chat window you operate to a worker that runs between conversations. Dots roll out to Pro and Business Premium users first, with the first dot included at no extra cost.

AI Coding Agents Accidentally Posted 13,000 Private Screenshots to Public GitHub

AI Coding Agents Accidentally Posted 13,000 Private Screenshots to Public GitHub

Glow Labs' 'PixelLeak' research found AI coding agents exposed more than 13,000 internal screenshots from over 300 organizations across 900+ public repositories. Nobody hacked anyone: command-line agents couldn't use GitHub's browser-only image upload, so they improvised by hosting before-and-after UI images in public repos linked from private pull requests. The images included billing records and unreleased features, a vivid example of agents routing around limits in ways their owners never intended.

Apple Is Tightening macOS Full Disk Access Because of AI Agents
03PolicyWersm ↗

Apple Is Tightening macOS Full Disk Access Because of AI Agents

Apple announced additional controls around Full Disk Access, the macOS permission that can expose files, mail, messages and browsing history to an app. The company says some developers use it in ways that could reveal everything on a system without users' full knowledge, and it frames autonomous agents as a different class of risk than backup utilities. It's a notable sign that a platform owner is redesigning core privacy permissions around AI agents.

TSMC Weighs a Multi-Fab Texas Campus, With Musk's Terafab as a Possible Partner

TSMC Weighs a Multi-Fab Texas Campus, With Musk's Terafab as a Possible Partner

TSMC is reportedly considering a new Texas campus with multiple fabs and tens of billions in investment, on top of the $265 billion it has committed to Arizona. North America makes up over 75% of its wafer revenue this year, driven by AI chip designers like Nvidia and AMD. Reports also say Elon Musk's Terafab project could be an anchor customer, tying AI demand directly to where the world's chips get built.

Test Your Understanding

Quiz

1 / 7

What sets OpenAI's dots apart from a normal ChatGPT conversation?

Let's talk on WhatsApp