Don't Assume Model Limits: AI Agent Hacks macOS Accessibility to Track Friend's Location
maxsloef · x · 2026-08-14
The author encourages users to constantly test LLMs on tasks they are sure the models cannot do yet, to truly understand where the frontier lies.
He quotes a surprising case: a user jokingly asked Codex to check if a friend was at a bar, expecting it to fail. Instead, the AI agent remotely opened the "Find My" app on a Mac Mini, used macOS accessibility hierarchy labels to locate the friend in the sidebar, and opened their popover to confirm their location.
More from coding & agent
- Developer Reports DSH Framework Shows High Efficiency in Complex Projects — ChrisGPT · 2026-08-14
- Entire drops waitlist: Git hosting for AI coding agents now open to all — craigsdennis · 2026-08-14
- Prime Agent Open-Sourced: A Recursive Agent That Rewrites Its Own Prompts — JeremyCMorgan · 2026-08-14
- Toast 1 Search Agent Tested: Frontier Quality at 1/10th the Price — xeophon · 2026-08-14
- AST-Retrieval MCP Saves 2.1M Tokens in Real Claude Code Roblox Project — naruto_uzumaki00 · 2026-08-14
- Sentry Launches Junior Agent: Ask in Slack, It Digs and Fixes Issues Automatically — zeeg · 2026-08-14