Commentary: Altman's Opaque AI Is creating a new security dilemma

Artificial intelligence researchers have been exploring a controversial idea: releasing less information about how AI models operate to make them more powerful and cheaper to run. However, this approach could make it harder to ensure the systems are not causing harm, potentially sparking a dangerous AI arms race. OpenAI CEO Sam Altman has embraced this risky strategy, launching a new model called Astra that reportedly utilizes a technique known as "recurrent depth."

This method can improve efficiency by not explicitly detailing the reasoning process in human language, making it more challenging for researchers to monitor the system's actions. While OpenAI claims it is still monitoring Astra, the company's move aligns with a broader trend in the industry, where pressure to stay ahead in the AI market is pushing firms to release potentially risky technologies.

This situation mirrors the security dilemma seen in international relations, where nations build up their capabilities to stay safe, leading to an escalation of risks.