OpenAI unifies its two experiments in autonomy — Operator’s browser control and Deep Research’s long-horizon synthesis — into ChatGPT Agent: a single system with its own virtual computer that can navigate websites, run code, manipulate files, and deliver finished artifacts like spreadsheets and slide decks.

The pitch is task completion, not conversation: ask for a competitive analysis and receive the deck; ask for a Japan trip and receive the itinerary and the bookings queue. Pro, Plus, and Team users get it first, with a monthly budget of agent tasks.

Buried in the launch, a milestone of a different kind: OpenAI classifies the agent as “High capability” in the biological and chemical domain under its Preparedness Framework — the first product to trip that wire — and ships it with correspondingly stricter safeguards. The company that spent years being accused of safety theater is now voluntarily labeling its own product potentially dangerous, which is either progress or marketing, depending on your priors.