[Is Astra's release next week in jeopardy? OpenAI concerned about the new model's powerful cyberattack capabilities] According to monitoring by Beating, OpenAI's latest tests have revealed that the programming and networking capabilities of the soon-to-be-released Astra model have advanced too quickly, making it impossible to rule out the possibility of autonomously attacking real critical systems. Previously, GPT-5.6 Sol was only at a lower tier. Based on OpenAI's standards, at this tier, the model may no longer require human supervision to independently identify and exploit zero-day vulnerabilities in critical systems, and could even complete the entire process from target identification to attack design and execution on its own. OpenAI has already suspended parts of Astra's internal testing and tightened permissions for networking, tool invocation, and model weights. Next, the model will be handed over to government agencies and external security organizations for testing. Earlier leaks suggested Astra was scheduled for release next week, but this timeline now appears uncertain. Altman also stated that Astra is very powerful and will eventually be made available to everyone, but the risks associated with its networking capabilities still require additional time to address. [Original link]