OpenAI Pauses Astra Development as Cybersecurity Risks Are Assessed

OpenAI has warned that its upcoming AI model, Astra, may possess “critical cybersecurity capabilities” after strong performance was observed in preliminary evaluations. Some internal development was paused, while tighter safety controls were introduced. The model is being tested in an isolated environment with restricted network access and sandboxed execution. Under OpenAI’s framework, critical cyber capability can be reached when real-world software vulnerabilities are autonomously identified and exploited. Sam Altman said broader access is still being planned, but additional time is being taken to ensure safe deployment.

Alibaba is reportedly planning a revenue-sharing model for major commercial users of upcoming open-weight AI models, including Qwen3.8-Max. Smaller developers and researchers would continue to receive free access to model weights, while high-revenue enterprises could be required to sign commercial agreements and share earnings. The strategy is being viewed as a shift toward sustainable AI monetization as infrastructure costs rise. Similar licensing has been introduced by Moonshot AI for Kimi K3. Alibaba has not publicly confirmed the reported pricing plans.

South Korea has unveiled an investment plan worth at least $880 billion to expand its semiconductor and artificial intelligence industries. New chip production hubs, AI data centres, and robotics technology will be developed under the country’s Three Mega Projects. However, concerns have been raised by investors over massive AI spending, and technology shares have been pressured in recent trading sessions.