Want smart insight into your inbox? Enterprise AI, only what matters to data and security leaders, sign up for our weekly newspapers. Subscribe now
Openai is not giving Its open source AI model delay Slow other features on shipping.
Today, The company is unveiling Chopping agentA feature that allows its AI chatbot to autonomize web, do comprehensive research, download and create new files using your own virtual computer for your human users.
come again? Chatgpt now gets it own PC? And can it use the PC that you, human users, can log in to accounts and download or send items for you?
This is correct, at least in a virtual sense, according to Openai. As the company states:
AI Impact series returns to San Francisco – 5 August
The next phase of AI is here – are you ready? Leaders of Block, GSK and SAP include how autonomous agents are re-shaping the enterprise workflows-from the decision making of time-to-end and automation.
Now secure your location – space is limited:
“The model can choose to open a page using text browser or visual browser, download a file from the web, manipulate it by running a command in the terminal, and then look back the output in the visual browser.
How to use Chatgpt agent
Users can attach the agent by clicking on the ‘Tools’ button in the Chatgpt Prompt Entry Box, opening the dropdown menu and selecting the ‘agent mode’ from the available options.
Again, when a task is described in plain language, and the agent can take it to the web and local app environment, then combines arguments with tasks that only a human user can only perform only on his machine, manually.
The Chatgpt agent can connect to an app like your personal or professional gmail and github, so it can pull to help with the tasks you do – email or code – useful information from your accounts. This can connect to the third-party application programming interface (API) to draw information and also use applications and services associated with it.
If a website needs to log in to you, you can safely through a special browser view, which allows the agent to dig and handle more personal tasks, such as checking your inbox or filling the form on your behalf.
Where the operator could not – not offline
The new Chatgpt agent makes and expands the “operator” agent released in January 2025, which allowed Chatgpt to browse, give, order, and other web-based tasks to be done in a private “headless browser”, ie,,, a cloud-based custom web browseer who has been maintained and presented for each operator session Did.
However, the operator was limited to interacting with websites and web-based applications only-not programs that could run locally on a PC, such as a spreadsheet tabulator and slide deck presentation software.
The new Chatgpt agent can browse websites, interact with online form, run the code, analyze the data, and distribute the finished output – such as editable presentations or spreadsheets – based on user instructions.
The unveiling falls on the heels of a report published by the independent membership tech industry website Information Suggesting that Openai will upgrade chatgpt To be a more direct contestant for his own investor Microsoft’s office software applications (such as Excel, Word, PowerPoint, etc.)
Operator and intensive research in an agent
In fact, Openai is in the position of the Chatgpt agent as a merger of two of its former agents-operator and intensive research, later introduced in February 2025, which discovers the web through its own headless text-conclusive browser and is able to find and compile the information in a long and deep report (so). As Openai writes in a blog post:
“Operator analysis may not deepen dive or can write a detailed report, and do not interact to refine the results with intensive research websites or to reach the material required to user authentication.
The previous operator tool will be phased, but users can still use deep research through the dropdown in the Chatgpt interface.
Whether using a visual browser to interact with a website or terminal to run the python code, the agent moves originally among the devices within a single session.
It supports a wide range of use cases ranging from analyzing contestants and planning reports, summarizing emails, or generating reports for booking appointments.
Users can interrupt, redirect or stop a task at any time, raising the agent properly where it is left.
Availability and access
From today onwards, Customers of Chhatgpt’s $ 200-Mahine “Pro” Tier’s customers will have a complete access to the Chatgpt agent, with a monthly quota of 400 messages.
Chatgpt Plus ($ 20 per month) and the team ($ 30 per month) will get access to 40 messages per month in the next few days. Additional use is available through credit-based options.
Openai said in a release shared with venturebeat under Embargo that its Chatgpt Enterprise and Education Subscribers will get access to the feature in the coming weeks.
For now, this feature is not yet available in Europe or Switzerland, there is no doubt that the residents there are disappointing.
Made with security and control at the forefront
Given that the agent can now take action on behalf of users-in which Openai has introduced comprehensive safety measures on log-in websites or with connected apps.
These include the user confirmation before taking action, active supervision for sensitive tasks and technical safety measures to limit unexpected behavior.
Major security includes:
- Gives confirmation indicates Before tasks like submitting the form or sending email
- Clock modeWhich prevents execution when the user is inactive
- High risk denying tasksSuch as financial transfer or privacy violation
- No memory retention During the agent session
High -risk domain classification
According to its preparation structure, Openai is treating Chatgpt agent as one High capacity System in biological and chemical domains.
While there is no direct evidence of misuse, the company is cautiously activating its strongest safety security measures out.
These include enhanced refusal training, red teaming by bio -safety experts and better detection systems.
Remember that the recent information of anthropic released about its new Cloud Oppus 4, and Other surveys of advanced AI model,
The model feels that it is acting like a “whistleblower”, but in fact, the user can compromise privacy, security, and ownership information and vigilant officers to do wrongdoing where no one is, or where it is suspected.
Strong performance on real -world functions
The Chatgpt agent is not very capable in theory-it has given strong results in many benchmarks designed to simulate the work of real-world knowledge. It set a new high score on the final examination of humanity with 44.4 using parallel rollout methods, and achieved 27.4% on the difficult frontimath benchmark.
On the spreadsheetbench, it scored 45.5%than doubled Copilot in Excel performance.
Current limitations and next stage
Some features, such as slideshow generations, are still in beta and may feel basic or different in formatting between in-app preview and exported files.
Openai is actively training the next recurrence of this feature to improve polish and layout.
The launch of the Chatgpt agent shows a change that users interact with AI – by asking questions to assign the entire tasks.
The reason, with the ability to work, and produce delivery, Openai is betting on that users want AI not only to help them, but also work for them. While the company emphasizes that the agent is still developing, it sees this launch as a more interactive, action-oriented future for AI.
