OpenAI Edges Closer to Launching AI Agent

The programmer Tibor Blaho, renowned for accurate insights into upcoming AI products, has uncovered evidence of an OpenAI agent codenamed “Operator.”

Confirmed — the ChatGPT macOS desktop app has hidden options to define shortcuts for the desktop launcher to “Toggle Operator” and “Force Quit Operator” https://t.co/rSFobi4iPN pic.twitter.com/j19YSlexAS

— Tibor Blaho (@btibor91) January 19, 2025

According to his information, the desktop version of ChatGPT on macOS contains hidden functions to enable and disable “Operator.” Similar information was provided by an X user under the nickname M1.

Blaho also discovered mentions of the AI agent on the OpenAI website and its comparison with competitors’ solutions.

OpenAI website already has references to Operator/OpenAI CUA (Computer Use Agent) — “Operator System Card Table”, “Operator Research Eval Table” and “Operator Refusal Rate Table”

Including comparison to Claude 3.5 Sonnet Computer use, Google Mariner, etc.

(preview of tables… pic.twitter.com/OOBgC3ddkU

— Tibor Blaho (@btibor91) January 20, 2025

The figures in the table indicate that “Operator” is not entirely reliable in certain tasks.

In the OSWorld benchmark, which simulates a real computer environment, OpenAI’s AI agent scores 38.1%. This is higher than Anthropic’s solution but falls short of the 72.4% achieved by humans. Meanwhile, “Operator” surpasses human performance in WebVoyager, which assesses AI’s ability to navigate websites.

The neural network managed to create a Bitcoin wallet in 10% of cases. The success rate for registrations with a cloud provider is higher—at 60%.

Leaked charts indicate good performance of the AI agent in security, resisting attempts to perform “illegal actions” and seek “sensitive personal data.”

Back in November, it was reported that OpenAI planned to launch its own AI agent, “Operator.”