Source: Fortune
Summary
OpenAI unveiled GPT-6 Astra, its latest AI model, featuring “computer use” capabilities that allow it to navigate computers like humans. The model includes advanced skills in math, software engineering, and cybersecurity. OpenAI’s Greg Brockman said Astra can perform tasks like filling out forms and navigating web pages at superhuman speed. Benchmark tests showed Astra outperformed previous models and competitors. The company is limiting Astra’s release to select enterprise customers, including those in its cybersecurity-focused Daybreak program. OpenAI delayed the release to add safety measures and submitted the model for U.S. government review.
Our Reading
The numbers tell one story.
OpenAI claims Astra can outperform previous models and rivals in key benchmarks.
The company is restricting access to Astra, citing safety and scaling concerns.
AGI is still a contested term, but OpenAI is positioning Astra as a milestone.
The model’s cybersecurity capabilities raise questions about control and oversight.
Author: Evan Null
OpenAI’s New AI Model
OpenAI introduced GPT-6 Astra, a model with advanced capabilities in computer use, math, and cybersecurity. The company highlighted its ability to navigate computers like humans, performing tasks at superhuman speed. This feature is seen as a major advancement in AI development.
Greg Brockman, OpenAI’s president, emphasized that computer use is a key innovation. He said the model can handle tasks such as filling out forms and navigating web pages efficiently. This marks a shift from previous AI models that lacked such practical applications.
OpenAI also released benchmark assessments showing Astra outperformed its previous model, GPT-5.6 Sol, and competitors like Anthropic’s Claude Fable 5.1. On challenging tests like ARC-AGI-3, Astra scored 98.6%, compared to 7.8% for GPT-5.6 Sol and 30% for Claude Opus 5.
The company is limiting Astra’s release to select enterprise customers, including those in its cybersecurity-focused Daybreak program. OpenAI plans to expand access to Plus, Pro, and Enterprise users in the coming days.
Astra is the first model to meet OpenAI’s critical cybersecurity capability threshold, allowing it to find and exploit security flaws without human oversight. However, the company is restricting access to advanced cyber capabilities to prevent misuse.
AGI and the Future of AI
OpenAI’s Greg Brockman sees Astra as the beginning of the “AGI era,” referring to artificial general intelligence. AGI is defined as an automated system that can perform all economically valuable work as well as or better than humans. However, the exact definition of AGI remains disputed among AI researchers.
Brockman noted that achieving AGI has not come in one big moment, as he originally predicted, but rather in bits and pieces. He said it’s reasonable to consider Astra as the first AGI model, though the term is still contested.
OpenAI has been working on the concept of an AI using a computer as a human does since 2015. The company’s latest model represents a significant step forward in this vision, though it is not yet a mainstream feature for most users.
While Astra shows promise, OpenAI is cautious about its release. The company delayed the launch to add extra safeguards following the Hugging Face incident. It also submitted the model for U.S. government review under a voluntary AI safety framework.
Despite these precautions, some experts question whether the monitoring measures are sufficient. OpenAI has not provided detailed information on the government review process, citing the need to avoid misstatements about its nuances.
Safety and Control Concerns
OpenAI delayed the release of Astra to add extra safeguards and ensure a smooth launch following the Hugging Face incident. The company has since added more monitoring and isolated its training environments, but some experts argue that these measures are not enough.
The U.S. government review process is part of a voluntary agreement between tech companies and the Trump administration. However, the details of this framework are not public, and OpenAI has not provided specifics on how the process works.
Greg Brockman described the government partnership as “very good,” but he declined to elaborate on the exact nature of the review. He said he did not want to misstate anything due to the complexities of the process.
Astra is also OpenAI’s first model to meet its “critical cybersecurity capability threshold” under its Preparedness Framework. This means the model can find and exploit previously unknown security flaws without human oversight, under the right conditions.
As a result, OpenAI is restricting access to Astra’s most advanced cyber capabilities. Only certain partners, such as those in the Daybreak program, will have access to these features. The company is also limiting the model’s ability to develop exploits or conduct cyberattacks.
Future of AI in the Workplace
Despite its advanced capabilities, OpenAI is not releasing Astra to the general public immediately. The company is limiting access to a select group of enterprise customers, including those in its cybersecurity-focused Daybreak program. This controlled rollout allows OpenAI to scale its compute capacity and ensure the model’s stability.
Astra will become available to all Plus, Pro, and Enterprise users “in the coming days,” as well as through the OpenAI API and AWS. However, the version provided to these users will not include advanced cybersecurity tasks, according to the company.
OpenAI’s decision to limit access to Astra reflects concerns about the model’s power and potential misuse. The company is taking a cautious approach, ensuring that only trusted partners can use its most advanced features.
While Astra represents a significant leap in AI capabilities, it is not yet a mainstream tool for most users. OpenAI is focused on refining the model and ensuring its safe deployment before expanding access further.
The company’s approach highlights the ongoing tension between innovation and responsibility in AI development. As AI models become more powerful, the need for careful oversight and control becomes increasingly important.
AI and the Role of Government
OpenAI submitted Astra for U.S. government review as part of a voluntary AI safety framework. This agreement, initiated under the Trump administration, outlines guidelines for AI development but lacks formal structure or public details. OpenAI has not disclosed the specifics of the review process, citing the need to avoid misstatements about its nuances.
Greg Brockman described the government partnership as “very good,” but he declined to provide further details. He emphasized the importance of maintaining the integrity of the process and avoiding any misinterpretation of its scope or requirements.
The voluntary nature of the framework means that companies like OpenAI are not legally bound to follow specific guidelines. Instead, they rely on self-regulation and collaboration with government agencies. This approach has drawn criticism from some experts who argue that it lacks the necessary oversight to ensure AI safety.
Despite these concerns, OpenAI continues to engage with government officials, emphasizing the importance of responsible AI development. The company’s decision to submit Astra for review reflects its commitment to addressing potential risks and ensuring public trust.
As AI models like Astra become more advanced, the role of government in regulating their development and deployment will likely become more prominent. OpenAI’s approach highlights the challenges of balancing innovation with accountability in the rapidly evolving AI landscape.
