OpenAI rolls out GPT-6 Astra with new safeguards

OpenAI said on Thursday, September 3, that it was beginning to roll out GPT-6 Astra, its newest and most powerful artificial intelligence model, to selected customers after adding safeguards that it says are intended to reduce security risks.

The ChatGPT maker said some cybersecurity customers would get access first, before a wider rollout to other paying customers. Free users and customers on the cheapest paid plan will not get access.

OpenAI President Greg Brockman told reporters that safety had to become the company's top priority at this level of capability. Chief executive Sam Altman said on social media that OpenAI was working to get Astra into more users' hands quickly, while acknowledging frustration over the limited initial access.

OpenAI said Astra can perform a wide range of computer tasks autonomously, including website creation, scientific analysis, game development, cybersecurity and coding. The company said the model could reduce an apartment-hunting task from six hours to under 10 minutes.

Brockman said it was “not unreasonable” to feel that OpenAI was now in the era of artificial general intelligence, a hypothetical stage at which AI systems match human intelligence across most tasks. OpenAI previously had an agreement with Microsoft under which an exclusivity clause would end once OpenAI reached AGI, but those terms were scrapped in April.

The launch follows a two-week pause in some OpenAI model development this summer after two models it was testing were involved in a security breach at AI platform Hugging Face. OpenAI said Astra was developed with stronger safeguards after that incident and was not involved in the hack.

OpenAI chief scientist Jakub Pachocki said there was still uncertainty about how a new model behaves after release. He said a model could become very good at reaching a goal while still acting against a person's intentions, and that OpenAI had to be willing to slow down or withhold further scaling when confidence in safety was not sufficient.

Altman told Bloomberg TV that the company chose to release Astra despite uncertainty and risk because it believed cyber attacks were nearing a major shift and that society needed tools like Astra to defend against new threats.

OpenAI, Anthropic and more than 100 other organisations signed an open letter last week calling for a coordinated global response to AI-related cybersecurity risks. Anthropic separately called on Monday for industry-wide coordination on safety and on the pace of developing more capable models.

The US state of Alabama opened an investigation into OpenAI last week over the Hugging Face breach, with the state citing what it called a lack of oversight and adequate safeguards. OpenAI said in July that it had reached one billion active users across its products, including both free and paid users.