苹果收紧全盘访问权限，起因涉及 Meta Muse

Welsh ICP Conviction 🏴󠁧󠁢󠁷󠁬󠁳󠁿🏉 @ICPLEGEND1966原推发布于 2026.10.03

在 X 上打开原推

🚨🤖🔐♾️ $ICP — AI AGENTS ARE BECOMING MORE POWERFUL. PERMISSION ARCHITECTURE MUST BECOME STRONGER. Apple has announced additional controls around macOS Full Disk Access, explicitly warning that increasingly capable and autonomous AI agents increase the risks associated with giving applications such broad access. Apple says Full Disk Access can expose files, mail, messages and browsing history, and future access will require more explicit user action. And the timing is VERY interesting. The announcement follows a dispute involving Meta’s Muse AI agent. Tech writer Jason Aten says he explicitly declined to connect Messages, yet later discovered that more than 187,000 rows from his Messages database had been synced. When questioned, Muse reportedly told him it was seeing incoming notification previews. Meta disputes the central allegation. Meta says Messages integration is opt-in and requires both macOS Full Disk Access AND the Messages connector to be enabled. Meta also acknowledged that Muse’s explanation about notification previews was incorrect. So let’s separate allegation from confirmed fact. Apple has NOT said Meta’s explanation is false. But Apple HAS now publicly warned: AI AGENTS + EXTRAORDINARY SYSTEM ACCESS = GROWING PRIVACY AND SECURITY RISK. And this takes us straight back to the infrastructure question I’ve been raising around $ICP. The future cannot simply rely on telling increasingly autonomous AI: “Don’t access that.” “Don’t send that.” “Don’t execute that.” “Don’t expose that.” We need systems where authority can be PROGRAMMATICALLY RESTRICTED. ♾️ Cryptographic identities ♾️ Canister-level authorization ♾️ Explicit permissions ♾️ Isolated application logic ♾️ Verifiable execution ♾️ Governance-controlled authority where appropriate ♾️ On-chain state and application logic The goal isn’t to pretend $ICP makes AI risk disappear. It doesn’t. The goal is to build AI infrastructure where the agent receives ONLY the authority required for the task — and where sensitive operations can require additional authorization rather than depending entirely on instructions inside a prompt. THAT DISTINCTION IS MASSIVE. PROMPTS ARE INSTRUCTIONS. AUTHORIZATION IS CONTROL. As autonomous AI becomes capable of controlling computers, money, communications, infrastructure and applications, permission architecture could become one of the defining cybersecurity problems of the AI era. Apple’s warning makes that conversation even more important. 🤖 MORE INTELLIGENT AGENTS. 🔐 SMALLER PERMISSION BOUNDARIES. ♾️ STRONGER CRYPTOGRAPHIC AUTHORITY. ⚙️ VERIFIABLE INFRASTRUCTURE. AI IS GETTING MORE POWERFUL. THE INFRASTRUCTURE CONTROLLING IT MUST GET BETTER TOO. ♾️ INTERNET COMPUTER | $ICP @dfinity @dominic_w @PierreSamaties #ICP #InternetComputer #DFINITY #AI #AIAgents #Apple #Meta #Muse #CyberSecurity #Privacy #Canisters #InternetIdentity #CloudEngines #DigitalSovereignty #SovereignCloud 🏴 Welsh ICP Conviction @ICPLEGEND1966 ❤️ LIKE 🔁 REPOST & SHARE 💬 COMMENT ♾️ ICP Donation Address: 1e672d038cebc619d93186418fa98f6499dbdb9cfdfac54f366c61a4a4ee4362

这条是引用着另一条写的：x.com/coinbureau/status/2106274299308826979