OpenAI has enhanced its security measures to block attempts at extracting internal reasoning from its GPT-6.2 Astra and GPT-6.2 Sol models. According to The Decoder, these improvements have drastically lowered the success rate of such extraction attempts from 54 percent to just 0.2 percent in adaptive red-teaming tests.

In parallel, OpenAI is actively suspending thousands to tens of thousands of accounts each week that are suspected of participating in coordinated distillation campaigns aimed at bypassing these safeguards, The Decoder reported. This ongoing crackdown underscores the company’s commitment to protecting its AI’s proprietary reasoning processes.

For Japanese investors and traders, these developments highlight the growing importance of robust AI security in financial markets, where advanced models like GPT-6.2 are increasingly integrated into trading algorithms and decision-making tools.