Tin tức mới nhất trong ngày, cập nhật liên tục 24h. Từ nguồn báo chính thống như CNN, BBC, VOA, Inside Science,...

OpenAI has begun rolling out a new agent mode for ChatGPT that can take actions on a user's behalf, signaling a major step toward turning chatbots into full‑fledged digital assistants.

The feature, available to Pro, Plus and Team subscribers, allows ChatGPT to “think” and “act” using a built‑in virtual computer. Instead of only providing information, it can now handle complex, multi‑step tasks such as checking a user's calendar and summarizing meetings or planning and purchasing ingredients for a meal. In a demo, the agent handled wedding preparation instructions—suggesting outfits, booking hotels, and more.

Agent mode combines and extends OpenAI's existing Operator (web browsing) and Deep Research (report‑building) tools. It reflects a broader industry race, as rivals like Google's Gemini and Apple's next‑gen Siri pursue similar ambitions.

OpenAI says the agent is experimental and includes safeguards. Data access is restricted, some tasks (like sending emails) need user confirmation, and high‑risk actions (like bank transfers) are blocked. CEO Sam Altman cautioned users to avoid giving it sensitive personal data until the technology matures.

The move highlights tech giants' vision of replacing app‑hopping with AI agents that handle everyday digital tasks—but also underscores ongoing challenges with privacy, bias, and unpredictable model behavior.