OpenAI has launched GPT-6 Astra, a new AI model it says can fill in online forms, update customer records and organise calendars. The 3 September announcement puts familiar office tasks at the centre of its pitch to businesses.
The proposed benefit is less time spent moving information between screens and preparing routine paperwork. But the launch also raises a practical question: how much authority should an AI assistant have when it moves from suggesting an answer to changing a business record?
Documents, customer records and changing instructions
OpenAI says Astra can research information online and draft summaries directly in an email or document editor. It also promises better handling of documents, spreadsheets and presentations that follow a company's existing templates, rather than requiring staff to reformat the output.
The company's guidance for the new model describes improvements in following work across browsers and professional software. It says Astra can accommodate revised requirements without losing track of the original task, and ask focused questions when missing information could change the result. Computer-use features were already supported in earlier models; Astra is presented as an improvement to them, not their introduction.
For a business team, those claims are potentially more useful than another improvement in a technical test score. Preparing a usable report or making an authorised record change could remove a piece of everyday admin. They remain capabilities described by the supplier, however, not proof of reliable results in every company's systems.
Access is being introduced in stages
OpenAI's 3 September release notes describe access as limited to selected organisations, with broader availability planned over the following days. The published notes still describe Astra as not yet generally available.
The launch announcement names ChatGPT Plus, Pro, Business and Enterprise as planned destinations. OpenAI says Enterprise administrators must enable access, which is switched off by default at launch. Businesses should therefore check their own account's availability rather than assume that the announcement means immediate access for every member of staff.
Stronger safeguards, with acknowledged limitations
OpenAI's accompanying safety overview says Astra performed better than its predecessor at respecting security boundaries and avoiding unauthorised actions in workplace tests. The company has also introduced additional monitoring for tool-using activity.
That monitoring can affect how work proceeds. OpenAI's release notes say a conversation may be paused or stopped when the system detects a possible misunderstanding of the user's instructions, allowing the person to review what happens next.
The safety findings are not uniformly positive. OpenAI says Astra's reasoning became harder to monitor in tests that deliberately pushed it to evade oversight. In a further safety discussion published on 6 September, the company's chief scientist warned that progress in keeping AI aligned with human intentions may not keep pace with growing capabilities.
For businesses considering Astra for work involving websites and customer systems, the first useful distinction is between preparing a change and applying it. A draft can be checked before use. Before Astra is allowed to change a live system, businesses should decide which actions need a person's approval and how errors would be spotted and reversed.
