Google has introduced ARTEMIS, an open-source framework that allows AI agents to control a real Android smartphone. This system enables AI models like Claude Code, Codex, Antigravity, or OpenClaw to perform tasks such as navigating apps, typing text, or chaining multiple actions on behalf of the user. ARTEMIS is available under the Apache 2.0 license and was initially designed for testing Android software. It can be used with a physical Android device connected to a PC or with a virtual phone in an emulator. Once set up, ARTEMIS can observe the screen, open apps, press buttons, type text, or switch between applications. Originally, ARTEMIS was aimed at developers and testing teams working on Android applications. For instance, it can install a new app version, reproduce a bug, navigate through multiple screens, or automatically capture screenshots and logs when issues occur. Unlike simple automated scripts, ARTEMIS can analyze the interface and determine how to achieve a given task. Google has demonstrated ARTEMIS performing tasks like planning multiple routes in Google Maps and then switching to YouTube to play a song. Two modes are available: Flash, which prioritizes speed, and Pro, which includes more planning and verification steps. ARTEMIS is not the first system to enable AI interaction with smartphones. Other tools, such as Gemini and Claude, can already perform multi-step tasks on Android apps. However, ARTEMIS stands out by transforming a real Android device into a versatile tool accessible to various external AI agents through its MCP (Mobile Control Protocol) server. This feature allows different AI models, including Claude Code, Codex, Antigravity, and others, to use ARTEMIS as an additional tool and assign tasks directly to the smartphone. On its first launch, ARTEMIS installs an accessibility service on the Android device, which is visible in the settings and shown through a persistent notification. This service allows ARTEMIS to read the interface structure and interact with elements on the screen. Google claims a success rate of over 99% on AndroidWorld, a benchmark covering more than 100 multi-step Android tasks. However, this figure is self-reported and has not been independently verified. Additionally, ARTEMIS has faced a controversy regarding code attribution. Minitap, a startup, claims that parts of its open-source project were used in ARTEMIS, leading the ARTEMIS team to acknowledge this in its documentation. A separate security issue on GitHub also highlights a potential vulnerability involving command injection through ADB functions. Until this is resolved, ARTEMIS should be treated as an experimental tool for testing purposes rather than for use on personal devices. While ARTEMIS is primarily designed for Android testing, it highlights the growing ability of AI agents to interact with smartphones directly. With the MCP server, it shows how AI is expanding its reach, turning smartphones into interfaces that can be manipulated by external agents. This development marks a significant step in the integration of AI with mobile technology.