Apple Tightens Mac Disk Access Controls Over Growing AI Agent Risks

Apple Tightens Full Disk Access Controls on Mac

Apple announced this week that it's adding new controls around macOS's "Full Disk Access" permission — a setting that grants an app near-total access to a user's files, mail, messages, and even browsing history. According to Apple, the move is a direct response to the growing capabilities of AI agents: "As AI agents become increasingly capable and autonomous, the risks associated with this level of access will grow substantially."

The change follows a disputed claim that Meta's Muse agent read a user's private messages without clear permission (which Meta denies), along with a Wired report on a flaw in ChatGPT's Mac app that could have exposed sensitive data. Going forward, apps will require "very explicit user action" before being granted this extraordinary level of access.

Why it matters: For small and medium businesses and individual users beginning to adopt desktop AI agents (like Meta's Muse or OpenAI's Dots), this is a concrete platform-level step toward protecting privacy — and a good reminder to review exactly what permissions you're granting AI tools on your computer.

Source: TechCrunch, The Verge

Meta Open-Sources Muse Hardware — From Raspberry Pi to E-Ink Displays

Meanwhile, Meta released "Muse Gadgets," an open-source project (firmware plus a Linux SDK) that lets developers and hobbyists connect its Muse AI agent to custom hardware: Raspberry Pi computers, ESP32 boards, e-ink displays, or even an HDMI stick for a TV. The company already built its own example — a device called "Muse Home Link" that connects the agent to a home network — and gave away 5,000 free units to subscribers.

This move is part of a broader Meta strategy to establish Muse as more than a standard chatbot, including "Muse for Small Business" (a free tier connecting to Shopify, Dropbox, and Slack) and the new Meta Enterprise Platform business unit aimed at pushing its AI products to companies.

Source: TechCrunch

Guide: All the AI Agents You Can Manage Through Plain Text Messages

TechCrunch published a comprehensive roundup this week of a fast-growing category — AI agents that operate directly through text messages (iMessage, WhatsApp, RCS) without requiring a separate app. Notable tools include Instinct (which raised $1 billion at a $10 billion valuation), Caddy and Town (focused on calendar management and professional tasks — relevant for small businesses), Fambot and Ohai ("chief of staff" assistants for families), and Poke, the first agent approved by Apple for its Messages for Business platform.

The trend shows how AI agents are becoming more accessible — simply by sending a text, like messaging a human contact — lowering the barrier to entry for small businesses wanting to experiment with automation without investing in custom development or a dedicated app.

Source: TechCrunch

In Brief: Another OpenAI Safety Employee Departs

In a separate development, David Robinson, who for three and a half years led the writing of safety reports accompanying OpenAI's major product launches, resigned and published a pointed essay in The Atlantic arguing the company's culture is "broken" and that its "iterative deployment" approach guarantees periodic failures that grow in scale as systems become more capable. This follows a string of similar resignations and warnings over the past month. OpenAI responded that it continues to strengthen its safety measures.

Source: TechCrunch