OpenAI is reportedly on the verge of launching "Operator," an AI tool designed to autonomously perform tasks on users' computers, such as coding and travel booking. According to software engineer Tibor Blaho, recent updates to OpenAI's macOS ChatGPT client include hidden options for "Toggle Operator" and "Force Quit Operator," indicating the tool's impending release. Additionally, Blaho discovered references to Operator on OpenAI's website, including performance comparisons with other AI systems. These findings align with earlier reports suggesting a January 2025 launch for Operator.
Benchmark data suggests that while Operator shows promise, it isn't yet fully reliable across all tasks. For instance, in tests requiring the creation of a Bitcoin wallet, Operator succeeded only 10% of the time. Despite these challenges, the development of AI agents like Operator is gaining momentum, with companies such as Anthropic and Google also entering the space.
However, the deployment of these agents raises safety concerns. One leaked chart indicates that Operator performs well in safety evaluations, including tests designed to provoke illicit activities or seek sensitive personal data. Ensuring robust safety measures is a key focus for OpenAI, especially given criticisms about the balance between rapid product development and comprehensive safety protocols.