OpenAI has announced that it cannot rule out its future model Astra reaching a critical level of cyber capabilities under its internal risk assessment system, the Preparedness Framework. Preliminary tests showed notable progress in agentic programming and solving cybersecurity tasks. This threshold implies that the neural network would be capable, without human involvement, of finding and creating working zero-day vulnerabilities for protected real-world systems, or independently building and conducting attacks after receiving only a general formulation of the goal.