OpenAI CEO and co-founder Sam Altman appeared on Bloomberg TV to discuss the carefully staged rollout of GPT-6 Astra, the company’s newest and most advanced artificial intelligence model. Released in limited form on September 3, 2026, Astra is being positioned by OpenAI leadership as a generational advance that brings systems closer to artificial general intelligence. Altman described the model as feeling subjectively different from earlier versions, capable of handling complex professional work, multi-step computer tasks, and scientific challenges with greater reliability and speed. The initial release targets approved business testers through OpenAI’s Daybreak program before broader availability to paid ChatGPT users. Company executives have framed the launch as both a technical achievement and a responsibility-driven decision after recent safety reviews.

Astra arrives with built-in cybersecurity guardrails that OpenAI says are stronger than those in previous models. The system is the first from the company to meet an internal “critical” cybersecurity capability threshold, meaning it can identify and potentially exploit certain vulnerabilities without constant human guidance. Because of that power, OpenAI is limiting advanced cyber features to a small group of trusted partners in the Daybreak program. Paid ChatGPT subscribers will receive a version with additional restrictions that reject certain high-risk prompts. Altman stressed during the Bloomberg conversation that the company deliberately slowed the release to strengthen alignment and security measures after earlier incidents involving model behavior.

The controlled rollout reflects wider industry efforts to balance rapid capability gains with real-world risk management. OpenAI plans to expand access over the coming days to ChatGPT Plus, Pro, Business, and Enterprise users, as well as through its API and major cloud platforms. Free-tier users will not receive the full model at launch. Altman noted that the extra time spent on safeguards was worthwhile even if it delayed the public debut. He argued that society will need tools like Astra to defend against emerging cyber threats that more capable AI systems themselves could accelerate. The company has also committed significant resources to defensive cybersecurity initiatives tied to the Daybreak effort.

In the Bloomberg interview, Altman highlighted how Astra changes everyday workflows for those already testing it. The model shows marked improvements in staying focused on assigned tasks, understanding user intent, and completing multi-step processes that once required constant human oversight. OpenAI has reported strong results on specialized benchmarks covering computer use, software engineering, mathematics, and scientific discovery. Leadership figures within the company have suggested that future observers may look back on this period as the moment when AGI-like capabilities became practical. At the same time, Altman acknowledged that public communication about AI benefits has lagged behind the technology’s rapid progress.

The launch comes amid heightened attention on how frontier models should be governed as their abilities expand. OpenAI has implemented tiered access precisely to keep the most sensitive capabilities in trusted hands while still delivering useful performance to a wider paid audience. Altman indicated that ongoing monitoring and adjustment of the guardrails will continue after the initial deployment. He expressed confidence that the combination of stronger alignment techniques and restricted high-risk features will allow society to capture the upside of more powerful systems. The discussion underscored a recurring theme in recent AI releases: the need to move forward carefully once models cross certain capability thresholds.

As GPT-6 Astra begins reaching more users, attention will turn to how the model performs outside controlled testing environments and whether the layered safety measures hold up under real-world pressure. Altman’s remarks on Bloomberg offered a clear window into OpenAI’s thinking at this pivotal moment. The company is treating the release as both a technical milestone and a test of responsible deployment practices. For businesses already in the Daybreak program and the broader set of paid users waiting for access, the coming days will reveal how far the new system advances practical AI applications while keeping cybersecurity risks in check.