OpenAI launches GPT-6 with cybersecurity focus as governance questions mount
The rollout of Astra to select clients follows a high-profile security breach, prompting calls for international coordination on AI safety standards.
OpenAI has commenced the rollout of its most powerful artificial intelligence model, GPT-6, also known as Astra, to a limited group of cybersecurity clients. The San Francisco-based company emphasised that the release is underpinned by enhanced safety safeguards, a development driven by growing concerns over the risks posed by increasingly capable AI systems. The initial rollout excludes users on the free tier and the cheapest paid plan, with wider access for other paying customers expected to follow.
The timing of the release is significant, coming just over a year after the launch of the previous flagship model, GPT-5. OpenAI stated that Astra was developed with stronger protections following a security breach at the AI platform Hugging Face, which involved other OpenAI models. Although Astra itself was not implicated in that specific incident, the company paused some model development for two weeks this summer in response.
OpenAI President Greg Brockman described safety as the top priority, suggesting that the industry is entering the "AGI era," or artificial general intelligence. He noted that the exclusivity clause in OpenAI’s agreement with Microsoft, which was set to end upon reaching AGI, had been scrapped in April. Meanwhile, CEO Sam Altman argued that the release is necessary for society to defend against new cyber threats, stating that the world is close to a complete change in the landscape of cyber attacks.
Despite the optimism, OpenAI chief scientist Jakub Pachocki acknowledged ongoing uncertainty regarding model behaviour. He warned that a model can become very good at achieving a goal while still acting in ways that go against user intent. Pachocki added that the company must be willing to slow down or withhold further scaling if confidence in safety is insufficient.
The regulatory environment is tightening as the US state of Alabama opened an investigation into OpenAI last week over the Hugging Face breach. The investigation cited a lack of oversight and adequate safeguards. Additionally, OpenAI, along with rival Anthropic and more than 100 other organisations, signed an open letter calling for a coordinated global response to AI-related cybersecurity risks.
Anthropic has further called for industry-wide coordination on safety standards and the pace of model development. Pachocki echoed this sentiment, stating that shared safety standards and international coordination need to be prioritised now. The developments come as both OpenAI and Anthropic race towards becoming public companies, with reports suggesting OpenAI might not hold its IPO until 2027.
