OpenAI hits pause on new model but pushes ahead with ‘always‑on’ AI agents

On September 29, OpenAI confirmed that it is postponing the launch of its newest large language model over internal security concerns. Company researchers reportedly warned that the safeguards and red‑team testing around potential misuse—such as cyberattacks or assistance with creating biological threats—were not yet sufficient. Yet, at the firm’s developer event, CEO Sam Altman still unveiled “Dots,” an “always‑on” AI agent designed to continuously assist users with complex tasks, highlighting a striking contrast in its strategy.

The delay reflects mounting pressure on so‑called “frontier AI” labs to prove they can keep increasingly capable systems under control. Governments and regulators are pressing for clearer incident‑reporting, independent safety evaluations and stronger internal governance, especially after a series of security scares around advanced models. At the same time, OpenAI is betting that agentic AI—systems that can autonomously plan and execute multi‑step work—will be a major revenue driver in areas like software development and back‑office automation.

This mix of hitting the brakes on the core model while accelerating commercial agents captures the central dilemma for leading AI companies: how to maintain rapid innovation without fuelling unacceptable risks. All eyes are now on what extra safeguards OpenAI adds before the postponed model eventually ships, and how that timeline may be shaped by emerging AI rules in the US and abroad.