ChatGPT Agent, OpenAI’s most capable assistant yet comes with an Altman warning

OpenAI CEO Sam Altman has announced the launch of ChatGPT Agent, a new AI tool capable of performing complex digital tasks on behalf of users. The launch comes with a warning that it is an experimental and potentially risky step into the next frontier of artificial intelligence.

The new ChatGPT system allows the chatbot to autonomously execute tasks using a local computer, including browsing files, booking reservations, analyzing work data, and preparing presentations.

In a post on X, Altman described the development as a significant leap in AI autonomy: “It can think for a long time, use some tools, think some more, take some actions, think some more, etc.” 

Today we launched a new product called ChatGPT Agent.

Agent represents a new level of capability for AI systems and can accomplish some remarkable, complex tasks for you using its own computer. It combines the spirit of Deep Research and Operator, but is more powerful than that…

— Sam Altman (@sama) July 17, 2025

From deep research to full autonomy

In demos, ChatGPT Agent was shown organizing travel plans for a wedding, purchasing an outfit, and drafting workplace documents using live data.

But the launch, while high-profile, came with a heavy dose of caution. “I would explain this to my own family as cutting edge and experimental,” Altman wrote. “A chance to try the future, but not something I’d yet use for high-stakes uses or with a lot of personal information until we have a chance to study and improve it in the wild.”

The ChatGPT Agent builds on earlier OpenAI efforts like Operator, which could complete real-world tasks like placing phone calls or managing reservations, and Deep Research, which could autonomously analyze documents or online sources at length. 

But this new system, Altman says, is more powerful, blending those features with an agentic model capable of multi-step reasoning and self-directed action across digital environments.

In practical terms, Agent allows ChatGPT to work as a personal assistant capable of interacting with the user’s browser, apps, and files. In a workplace example, it can process large data sets from emails or spreadsheets and produce a summarized report with visual slides, executing each step independently.

Altman’s risk warnings

Despite the system’s capabilities, Altman was emphatic about the risks. “We don’t know exactly what the impacts are going to be,” Altman wrote, warning that bad actors may try to manipulate users’ agents in unpredictable ways. “We recommend giving agents the minimum access required to complete a task to reduce privacy and security risks.”

The company has built extensive safeguards, including user control over what the agent can access, session logs with replay features, and system-level boundaries to avoid dangerous actions. 

Still, Altman emphasized that the technology should be adopted slowly. He compared using it to “cutting edge experimental tech,” noting that while the Agent is safe for certain tasks, like finding time for dinner, it should not yet be trusted with unrestricted access to sensitive areas like email inboxes or financial documents.

With this release, OpenAI signals its long-term ambition to create digital agents that move well beyond conversational interfaces. In a competitive field where Microsoft’s Copilot, Google’s Gemini, and now X’s Grok are vying for the same space, ChatGPT Agent represents OpenAI’s clearest attempt yet to own the interface layer for digital productivity.

Altman stressed that society and institutions must co-evolve with the technology. “As with other new levels of capability, society, the technology, and the risk mitigation strategy will need to co-evolve,” he said.