AI spotted trouble on a compromised Mac
Ben Thompson describes an exposed screen-sharing service on a Mac set aside for an AI program that can take actions. He says a second AI program flagged suspicious files and stopped commands. That is one account, not a general security test or a reason to remove consent checks.
What the source reports
Ben Thompson writes about an attack on his Mac Mini. It was set aside for AI software that could take actions. An exposed screen-sharing service let the attacker in. Thompson says a second AI program spotted odd files. It then stopped its commands. That helped in this one case. It does not show how often AI can catch an attack. Thompson says standard permission prompts do not fit this work. The person may not be at the screen when AI asks to act. That problem needs a careful fix. It is not a reason to give software full access. Lock down remote access. Keep permissions narrow. Make it clear when a person must say yes. Keep a way to stop the work.
Original source
- Title
- Apple and a Hacker’s Future
- Author
- Ben Thompson
- Publication
- Stratechery
- Date
- Monday, October 5, 2026