OpenAI Launches Astra, Its Most Powerful Yet Controversial AI Model

On September 3, OpenAI released GPT-6 Astra, describing it as its most capable and best-aligned model to date, with improvements in computer operation, browser usage, software engineering, and complex specialized tasks. Company President Greg Brockman stated that Astra expands the range of work people can entrust to AI. Astra was first rolled out to enterprises participating in the Trusted Access Program and will gradually become available over the coming days to ChatGPT Plus, Pro, Business, and Enterprise users via API. Its state-of-the-art cybersecurity capabilities will not be fully opened simultaneously but will initially be used by a small group of alpha testers before being progressively provided to vetted defense agencies through Daybreak Blue. OpenAI said that Astra’s ability to identify and exploit zero-day vulnerabilities could help security teams discover and patch weaknesses, while noting that measures such as rejecting malicious requests, restricting unauthorized operations, and monitoring chain-of-thought processes have been added. Claims regarding Astra’s adoption of “opaque recurrence,” which may reduce the monitorability of reasoning, currently stem mainly from media reports. Chief Scientist Jakub Pachocki acknowledged that monitoring the reasoning process becomes more difficult as models complete tasks using fewer or no language tokens. Regarding whether Astra signifies that AGI has already been achieved, Brockman remarked that this is no longer a contractual trigger concept and personally believes we have reached that point, though this does not equate to an official announcement by OpenAI that AGI has been realized.