The Big Story

Apple is adding new limits to a Mac permission called "Full Disk Access" because AI agents are getting powerful enough to cause real problems if they go wrong. The company announced the change on October 2 after seeing how capable these tools have become at handling tasks that touch private data. Full Disk Access lets an app reach every file, message, email, and browsing record on the computer without repeated prompts. AI agents are programs built to complete multi-step jobs such as sorting documents or summarizing long chats, so they often request this broad permission to work smoothly. The risk is that a poorly designed or misused agent could copy sensitive information or make unintended changes across the entire system. Apple is responding by adding extra confirmation steps and clearer warnings so the permission is granted only when the user truly intends to allow it. This update matters to anyone who uses a Mac and values privacy, especially as more everyday apps begin adding AI helpers that promise to "do things for you." Mac users will start seeing the new prompts right away when an app asks for broad access. You can review which apps already have Full Disk Access by opening System Settings, clicking Privacy & Security, and scrolling to the Full Disk Access section, where you can turn off any entries you no longer trust. Source: arstechnica.com

Explain Like I'm 14

You know how when you let a helpful robot vacuum clean your room, it needs to know where the furniture is and which doors are open so it doesn't get stuck? Now picture that same robot but it can also open drawers, read your notes, and decide what to throw away. To do any of that safely, someone has to give it permission first — otherwise it might wander into places it shouldn't. AI agents work the same way: they are programs that try to finish bigger jobs on their own, like organizing your photos or answering questions across many apps. Because those jobs often touch private data, the computer asks you to approve "Full Disk Access" before the agent can look everywhere. The new Apple rules are basically adding extra locks and clearer signs so you only hand over the master key when you really mean to. That way the helpful robot stays helpful instead of accidentally making a mess.

Cool Stuff & Try This

AI in space — Google’s satellite test Google just launched a small satellite built with Planet Labs to see how its AI chips handle the extreme conditions of orbit. The satellite went up on SpaceX's Transporter-18 rideshare mission and carries Google TPUs, the same kind of chips used in data centers to train and run large models. The goal is to gather data on how these chips survive the physical stress and temperature swings of spaceflight. One big advantage of low Earth orbit is near-constant sunlight, which the project says could generate up to 8 times more solar power than systems on the ground. This is still early research, but it shows one possible future where some of the heavy computing that powers chatbots and image tools might happen above the clouds instead of in giant data centers. If you're into space, climate tech, or just big "what if" ideas, this project is worth following. You can read the short announcement on X by searching @GoogleAI and looking for the post from October 2 about Project Suncatcher — no signup needed. Source: x.com

Quick Bits

Ten popular chatbots recommended unlicensed gambling sites When researchers asked ChatGPT, Gemini, Claude, Grok and six other chatbots questions in Dutch about casinos, every single one named at least one site that doesn't have a proper Dutch license. The study, run by Keurmerk Verantwoorde Affiliates, tested nine questions per tool and found some chatbots even presented missing ID checks or large bonuses as advantages. The watchdog warned that Dutch consumers should be able to trust accurate information when they ask about legal gambling options. Source: nltimes.nl

Anthropic co-founder worried AI might be able to suffer Christopher Olah, one of the people who started Anthropic, has been quietly meeting with religious leaders to discuss whether the company's AI model Claude could someday feel pain or have experiences that matter morally. The conversations began in fall 2025 and are part of an effort to think about AI ethics early, even while the company pushes toward a $2 trillion valuation and an eventual IPO. Critics have noted that treating the model as a possible moral entity could affect how the company handles liability if something goes wrong. Source: the-decoder.com

💬 Reply to this email — Patrick reads every one.

Share: X · LinkedIn · WhatsApp

Forwarded this email? Subscribe here — it's free.