OpenAI launches GPT-6 Astra

Uses a computer independently and reaches the highest level of cyber risk

Advanced artificial intelligence capabilities are no longer limited to answering questions or generating text. On Thursday, OpenAI began rolling out the sixth generation of ChatGPT, capable of independently using a computer and executing complex tasks, alongside a monitoring system designed to halt the model if it exceeds its predefined boundaries.

The launch of GPT-6 Astra comes a year after the release of its predecessor, following a series of cyberattacks carried out during the summer using AI tools that were under testing, including technologies developed by OpenAI and its competitor Anthropic. These incidents highlighted the risks associated with highly advanced models.

Browse Maps

Cultural News

Politics

The company plans to make the new version available initially to a "limited number of organizations," before releasing it in the "coming days" to subscribers of paid plans and developers.

In a demonstration, OpenAI showcased the model receiving voice commands and independently executing a task to post a sales advertisement.

According to research results published by the company, the model can find housing within 10 minutes—a task that might take a specialized employee around six hours.

GPT-6 is the first model developed by OpenAI to reach the highest risk level according to its cybersecurity assessment framework, as it can breach highly protected information systems without assistance.

Due to these capabilities, the company has restricted the most sensitive functions to cybersecurity specialists, relying on an automated monitoring system capable of stopping user tasks when necessary, "including in operations unrelated to cybersecurity," according to the company.