OpenAI, the company behind ChatGPT, announced on 3rd September that it would start deploying its latest and most advanced artificial intelligence (AI) model to its select customers, noting that it had implemented security measures to prevent security risks.

“At this level of capability, safety has to be our first priority,” OpenAI President Greg Brockman told the reporters on a conference call about the release of GPT-6, aka Astra.

Just over a year has passed since OpenAI launched the previous version of its flagship model, GPT-5.

But since then, worries over the skill of the more sophisticated systems have increased, after failures of systems developed by OpenAI and another developer, Anthropic.

This summer, OpenAI suspended some of its model development for two weeks due to the security breach of two models it was testing on the AI platform Hugging Face.

Following that incident, the San Francisco-based firm said all the new security measures are in place for the development of Astra, which wasn’t involved in the hack itself.

The new model will be available to some cybersecurity customers on Thursday, with a broad rollout to other customers who pay for cybersecurity services following, the company said. Users in the free tier or the lowest-priced paid subscription won’t have access.

“We are working towards getting Astra in everyone’s hands as quickly as we can; I know it is frustrating and I appreciate the patience. It should be quick,” OpenAI CEO Sam Altman posted on social media Thursday afternoon.

In a blog post, OpenAI stated that the AI can complete a diverse set of “tedious” computer-related tasks without human supervision, such as building websites, analysing scientific information, creating games, cybersecurity and coding.

The company provided an example of how the model could speed up workers’ time spent researching apartments from six hours to less than 10 minutes, illustrating the time savings of building autonomous AI agents with Astra.

“It’s not inconceivable to think that we’re in the AGI era now,”  Brockman said in the call, meaning the artificial general intelligence phase in which AI systems are capable of performing most tasks as well as humans.

There was a prior deal with an early and big investor, Microsoft, that included an exclusivity clause that would expire when OpenAI achieves AGI. Those terms went out of use in April.

‘Limited window’

OpenAI’s chief scientist Jakub Pachocki said there was still speculation about the behaviour of a new model once it is released.

“A model can be very effective at achieving a goal and yet can do things that the person doesn’t mean it to do,” Pachocki said on the same call.

“We must also be ready to scale back or cease further scaling when our confidence with safety isn’t high enough,” he continued.

In the Bloomberg TV interview Thursday afternoon, Altman spoke about what made the company choose this amount of uncertainty and risk to go ahead with a new model release.

“It’s a new world, there’s a huge shift in the way cyber attacks are going to be conducted, and the only way I see society collectively protecting itself from these new cyber threats is by using tools like Astra to quickly defend against them,” Altman said.

In July, OpenAI announced it had surpassed one billion active users, encompassing both free and paid users on all its products.

Last week, over 100 organisations, including OpenAI, Anthropic, and other biased counterparts, called for a coordinated global effort to address the cybersecurity threats posed by AI, stating that they had only a narrow window to enhance cyber defences.

Anthropic took one step further on Monday, urging the industry to coordinate on the pace of increasingly capable models and on safety.

The shared safety standards and international coordination on further AI development should be a priority now, Pachocki said on Thursday at OpenAI.

His office stated in the filing that it was concerned with OpenAI’s “total lack of oversight and adequate safeguards” regarding the Hugging Face hacking incident, prompting the US state of Alabama to initiate an investigation into the company last week.

It also comes as OpenAI and rival Anthropic are both racing towards becoming public companies over the next several months, though OpenAI might not hold its IPO till sometime in 2027, according to reports.