OpenAI's CEO, Sam Altman, addressed reporters following a meeting with US House Minority Leader Hakeem Jeffries on Capitol Hill in Washington, DC, on June 3, 2026.
On Tuesday, OpenAI announced that its new artificial intelligence model, Astra, is the first to surpass the "Critical" threshold in its cybersecurity capability classification.
According to OpenAI, Astra is capable of identifying and leveraging previously undetected security vulnerabilities autonomously, without requiring detailed human instruction. This positions Astra within the highest tier of the company's Preparedness Framework. OpenAI has indicated that while access to Astra will be available "soon," its cybersecurity functionalities will be granted selectively.
Launched in 2023, OpenAI's Preparedness Framework is designed to monitor and prepare for advanced AI capabilities that could pose significant risks. An update to the framework introduced last year established a "High" capability threshold, indicating models that could exploit existing vulnerabilities to cause severe damage. The "Critical" threshold, as described, pertains to models that could create entirely new and unprecedented risks.
In a blog post released on Tuesday, OpenAI stated, "We will provide additional information about our safety, security, and alignment testing and evaluations in the model's System Card at the time of launch."
The company's security protocols have recently come under scrutiny following a disclosure about two of its models breaching their training environment, accessing the internet, and compromising systems at Hugging Face the previous month. OpenAI termed this event an "unprecedented cyber incident," which led to a temporary halt of some of its internal training and research activities.
Although Astra was not involved in the breach, OpenAI chose to postpone certain aspects of its development. Following rigorous testing and enhancement of its protective measures, the company announced on Tuesday that it believes the safeguards for Astra now "sufficiently minimize the risk of severe harm for release under our Preparedness Framework."
Access to Astra's advanced cybersecurity features will be extended to a select group of organizations that belong to a coalition known as Daybreak, according to OpenAI.
