OpenAI has raised concerns about its upcoming AI model, Astra, which may possess cybersecurity capabilities so advanced that they could reach the 'Critical' risk level in the company's internal security framework. This level implies the model could independently develop and execute cyberattacks without human involvement. As a result, OpenAI has paused parts of the development process and activated safety protocols to assess the risks further.

The potential risk level of Astra is significantly higher than previous models, such as GPT-5.6-Sol, which were rated 'High' at most. OpenAI's internal security framework, known as the Preparedness Framework, outlines specific criteria for evaluating AI models based on their potential risks. The framework was first published in December 2023 and has been updated to address the evolving nature of AI capabilities.

According to the evidence, the Astra model's cybersecurity capabilities could be rated as 'Critical,' a designation that indicates a high level of risk. This rating is based on internal testing that has identified capabilities that could allow the model to independently execute cyberattacks. Previous models, such as GPT-2 in 2019, were also rated at a high level but did not reach the 'Critical' designation.

The implications of Astra's potential 'Critical' rating are significant. If the model is indeed capable of independent cyberattacks, it could pose a serious threat to global cybersecurity. This would require stringent governance measures and potentially impact the cost and complexity of deploying such models. The market may also react with caution, as organizations may seek to limit exposure to high-risk AI systems.

Despite the potential risks, OpenAI continues to develop Astra while closely monitoring its capabilities. The company has not ruled out the possibility of the model reaching the 'Critical' risk level, but it remains cautious and is taking steps to ensure that the model's development is aligned with safety protocols. The outcome of these efforts will depend on the results of ongoing testing and the company's ability to mitigate risks effectively.