OpenAI has shipped a new integration that lets the ChatGPT macOS desktop app connect directly to Apple's Messages app on Apple Silicon Macs. Once a user grants the appropriate system permissions, the assistant can read incoming iMessage, SMS, and RCS conversations, search across message history, summarize long threads, draft replies in a user's own voice, and, with explicit confirmation, send messages on the user's behalf. The feature works inside ChatGPT's Work and Codex modes and is rolling out across all subscription tiers, including the free tier, though it is limited to Apple Silicon hardware and does not support Intel-based Macs. For developers and product teams, this matters less as a consumer convenience feature and more as another data point in the broader shift toward agents that act inside a user's existing tools rather than requiring people to copy-paste context into a chat window. It follows a pattern seen across the industry this year: assistants are increasingly granted narrow, permissioned access to native OS surfaces, including calendars, file systems, browsers, and now messaging apps, so they can complete multi-step tasks autonomously. That raises real engineering and security questions builders should track: how permission scopes are audited, what telemetry OpenAI collects when an agent reads private message content, how reversible or logged "send on my behalf" actions are, and whether third-party developers will eventually get API-level access to equivalent Messages hooks rather than this remaining an OpenAI-exclusive integration built on private Apple frameworks. Teams building competing AI agents or Mac-native automation tools should note the precedent this sets for what Apple is willing to expose to outside AI vendors, since Apple has historically restricted deep Messages access even for its own ecosystem partners. Expect scrutiny over data retention and whether message content is used for model training.