Oct 5, 2026 · 19 min
Mac automation turns security strength into an agent-hosting risk
Apple and a Hacker’s Future
As AI agents gain access to powerful computers, macOS shows how automation and accessibility can expand both capability and the consequences of compromise.
- 1A compromised Mac mini used for AI agents was detected mining cryptocurrency, exposing the risks of always-on agent hosts.
- 2macOS offers unusually powerful automation through Unix access, accessibility APIs, and scripting, but those same tools widen the attack surface.
- 3Apple’s smart-home ambitions contrast with more open, hacker-oriented projects that favor experimentation over packaged convenience.
Don't miss
Thompson discovers that his always-on Mac mini, running AI agents, has been compromised and is mining cryptocurrency.
The brief
A high-severity macOS screen-sharing vulnerability sets the stakes: internet-exposed port 5900 can expose systems, making remote access a security decision rather than a convenience.
Ben Thompson recounts finding that his always-on Mac mini, running Claude and Codex, had been compromised and was mining cryptocurrency; an agent stopped commands and helped reconstruct the intrusion.
The episode argues that macOS is an unusually capable home for computer-using agents because Unix, scripting, accessibility APIs, and automation are deeply integrated—but compromise carries equally broad consequences.
Apple’s permission model adds friction without eliminating risk: Full Disk Access and TCC prompts can expose sensitive data, while safer VPN or SSH workflows complicate everyday screen sharing.
The discussion closes on Apple’s smart-home push and Muse Gadgets, contrasting a packaged assistant strategy with open-source hardware and the Mac’s older hacker appeal.