OpenAI may be on the verge of releasing its AI tool, Operator, which can take control of PCs to perform tasks autonomously. Reports suggest a target release date in January.
Details on OpenAI’s Operator tool
Tibor Blaho, a software engineer known for accurately leaking information about upcoming AI products, has reportedly found evidence of the long-anticipated Operator tool. Bloomberg and other publications have previously discussed Operator as an “agentic” system capable of handling tasks such as writing code and booking travel.
According to The Information, OpenAI is aiming to launch Operator in January. Blaho’s recent findings, including code, bolster this timeline.
The ChatGPT client for macOS has received updates, which include hidden options to set shortcuts for “Toggle Operator” and “Force Quit Operator,” as reported by Blaho. Additionally, references to Operator have been observed on OpenAI’s website, although they are not yet visible to the public.
Blaho also noted that OpenAI’s website features tables comparing Operator’s performance to other computer-using AI systems, which may be placeholders. If the data in these tables are accurate, they indicate that Operator has reliability issues, varying by task.
In benchmarks conducted by OSWorld, labeled “OpenAI Computer Use Agent (CUA)”—presumed to be the model supporting Operator—scored 38.1%. While this score exceeds that of Anthropic’s computer-controlling model, it remains significantly below the human score of 72.4%. The OpenAI CUA surpasses human performance on WebVoyager, a benchmark assessing an AI’s website navigation skills but does not achieve human-level scores on another benchmark, WebArena, according to the leaked data.
Operator has demonstrated challenges in executing tasks that humans can easily perform. In a trial involving signing up with a cloud provider and launching a virtual machine, Operator succeeded just 60% of the time. In another task of creating a Bitcoin wallet, Operator was only successful 10% of the time, based on the leaked information.
OpenAI’s impending introduction of Operator aligns with efforts from competitors such as Anthropic and Google, who are also investing in AI agents. Despite the risks and speculative nature of AI agents, tech giants promote them as the future of AI technology. Markets and Markets, an analytics firm, forecasts the AI agent market could reach a valuation of $47.1 billion by 2030.
Current AI agents are considered primitive, yet there are concerns over safety should the technology advance quickly. One leaked chart indicates Operator performed well in specific safety evaluations, including tests designed to deter it from executing “illicit activities” or searching for “sensitive personal data.” The focus on safety is noted as a contributing factor to Operator’s protracted development cycle.
In a recent post on X, OpenAI co-founder Wojciech Zaremba criticized Anthropic for launching an agent that reportedly lacks safety mitigations, stating, “I can only imagine the negative reactions if OpenAI made a similar release.”
OpenAI has faced criticism from AI researchers, including former staff, for allegedly prioritizing rapid product rollout over safety considerations.
Featured image credit: Tibor Blaho
