OpenAI begins phased rollout of GPT-6 Astra model

San Francisco-based artificial intelligence company OpenAI announced on Thursday that it will begin rolling out its latest and most advanced AI model, GPT-6, also known as Astra, to select customers. The company stated that it had implemented safeguards to address security risks associated with the model's unprecedented capabilities.

OpenAI president Greg Brockman emphasized the need to prioritize safety at this level of AI development. Concerns about advanced AI models' capabilities have intensified since the previous version, GPT-5, was released a year ago. Following incidents involving OpenAI and rival developer Anthropic, OpenAI temporarily halted model development for two weeks this summer after two models were involved in a security breach at AI platform Hugging Face.

Astra has been designed with strengthened safeguards, though it was not implicated in the Hugging Face incident. Some cybersecurity-focused customers gained access to Astra on Thursday, with a broader rollout planned for other paying customers. Users on the free tier or the most affordable paid plan will not have access. OpenAI described Astra's ability to autonomously manage various computer tasks, such as website creation, scientific analysis, game development, cybersecurity, and coding.

The model could potentially reduce apartment hunting from six hours to under ten minutes. OpenAI's chief scientist, Jakub Pachocki, acknowledged the uncertainty surrounding the model's behavior post-release, stating that models can still act against user intentions even if they excel at a specific goal. Pachocki added that OpenAI is prepared to slow down or halt further scaling if confidence in safety is insufficient.

The artificial general intelligence (AGI) era may be upon us, according to Brockman, referring to a hypothetical stage where AI systems match human intelligence across most tasks. OpenAI previously had an agreement with Microsoft, a significant investor, which would end exclusivity once AGI was reached, but this was abandoned in April.

A limited window exists for implementing safety measures, as acknowledged by OpenIA's chief scientist. OpenAI, Anthropic, and over 100 other organizations recently signed an open letter urging a coordinated global response to AI-related cybersecurity risks, warning that the window to enhance cyber defenses is limited. Anthropic took a stronger stance on Monday, calling for industry-wide safety coordination and a slower pace of developing increasingly capable models.

OpenAI's chief scientist emphasized the need for shared safety standards and international coordination in AI development. The US state of Alabama has opened an investigation into OpenAI following the Hugging Face breach, citing the company's perceived lack of oversight and inadequate safeguards.