Apple and a Hacker's Future

Your Mac may be vulnerable to remote hacking if its screen sharing feature has been left on with an exposed port.

A critical macOS vulnerability, CVE-2026-65400, allowing remote code execution and unauthorized access, has been actively exploited. The vulnerability, stemming from a flaw in macOS screen sharing, enabled attackers to gain root access and install crypto miners. Apple released a patch for macOS Tahoe, Sequoia, and Sonoma last week. The details of CVE-2026-65400 became public at the Black Hat security conference. The Netherlands National Cyber Security Centrum (NCSC) warned of active exploitation on systems with port 5900 accessible from the internet. This incident highlights the risks associated with persistent agent-based computing, even as the author utilized an AI agent to detect and help remediate the breach before discovering the vulnerability's details.

Apple's recent disclosure regarding "Full Disk Access" indicates a significant shift in how it will manage application permissions, especially for AI agents. The company plans to introduce stricter controls, requiring explicit user action for granting extraordinary access levels. This move is driven by the growing autonomy and capability of AI agents, which Apple believes pose substantial risks to user data and privacy if such broad permissions are not carefully managed.

The author's experience underscores the tension between Apple's security model, particularly the Transparency, Consent, and Control (TCC) subsystem, and the operational needs of advanced AI agents. TCC's GUI-only permission prompts are invisible to headless systems and agents, leading to silent failures and operational headaches. While TCC aims to protect users, it creates a significant barrier for users who rely on automated agents for tasks on their Macs, forcing workarounds that can inadvertently introduce security risks.