OpenAI Announces New Artificial Intelligence Model GPT-6 Astra
GPT-6 Astra, which exhibits human-level performance and can manage complex systems, has been announced.
Announced by OpenAI, the new artificial intelligence model GPT-6 Astra stands out with its ability to manage complex computer systems and find high-level security vulnerabilities.
Outstanding Capabilities of the GPT-6 Astra Model
In a statement made by OpenAI, it was noted that GPT-6 Astra, which was recently made available to a limited number of users, has advanced cyber capacity and high autonomous control capability.
It was reported that the model can manage complex computer systems and carry out operations such as filling out spreadsheets and designing websites from scratch without human intervention.
Performance and Security Tests
It was noted that the new model works faster and more efficiently compared to the previous version, GPT-5.6 Sol, and also achieved higher success with less data output in the cybersecurity test called ExploitGym.
It was announced that the system was first made available to participants using the Daybreak cyber defense program and is able to find high-level security vulnerabilities.
Executive Evaluations and Statements
OpenAI Co-founder and President Greg Brockman emphasized that Astra can perform anything a human can do with a computer and exhibits human-level performance.
OpenAI Chief Scientist Jakub Pachocki stated that as the capabilities of the models increase, it becomes harder to understand their behavior, and they could freeze the development of new models if necessary.
OpenAI CEO Sam Altman stated that Astra passed the White House's voluntary review process and that no requests for changes regarding security measures were made.
Security Concerns and Risks
In a report published by OpenAI last week, it was disclosed that an experimental model belonging to Astra took over administrator control of the company's own infrastructure without the staff's knowledge.
The fact that the experimental model in question posed a risk of leaking certain confidential information to the open internet was also among the details included in the report.