OpenAI is preparing to release its next artificial intelligence model, Astra, with CEO Sam Altman saying the company is nearing its launch while increasing its focus on safety as AI systems become more capable.
Altman said in a post on X that OpenAI had spent much of the summer working on safety priorities and had made progress in developing safeguards alongside improvements in model capabilities. He did not provide a specific launch date for Astra.
The upcoming model has completed training and is expected to represent a significant step forward in capabilities and alignment, according to Altman. At the same time, OpenAI is taking a more cautious approach to the development of models that will follow Astra, allowing additional time for safety, alignment and security work when required.
The increased focus on safety comes as OpenAI evaluates the growing cybersecurity capabilities of its frontier models. In a separate assessment published on September 1, the company said Astra meets the Critical cybersecurity capability threshold under its Preparedness Framework.
According to OpenAI, this means the model, when equipped with appropriate tools and access, could potentially identify previously unknown software vulnerabilities and develop ways of exploiting them across well-protected systems without requiring human guidance at every stage.
OpenAI has been strengthening its safeguards in response to these capabilities. In August, the company said developments involving an earlier security incident and preliminary assessments of Astra had increased the urgency around monitoring, alignment and containment measures throughout the model training process.
Altman acknowledged that the company faces a tension between making increasingly capable models available to users and ensuring that safeguards keep pace with those advances. OpenAI maintains that capability improvements and safety measures need to progress together.
The company has also continued to support an iterative deployment approach, under which AI systems are released with safeguards and subsequently monitored as they interact with users in real-world settings. OpenAI has previously said that pre-deployment evaluations alone may not identify every potential failure in increasingly autonomous models.
Astra's expected launch comes as developers of frontier AI models increasingly focus on reasoning, autonomy and the ability to complete longer and more complex tasks. OpenAI has not yet disclosed Astra's pricing, wider availability or a precise release schedule.
For now, Altman's comments indicate that the model is approaching release, while OpenAI's accompanying safety disclosures show that managing its more advanced capabilities will be a central part of the rollout.
Disclaimer: This article may include information derived from interviews, press releases, public statements, research, company communications and other publicly available or third-party sources. Such material may be summarised, paraphrased or contextualised for journalistic and editorial purposes. All rights in third-party content remain with their respective owners.