Industry warning: prompt injection risks and models' resilience

The author of the post reports that OpenAI's new model is roughly equivalent to Gemini Flash and Opus 4.8 in terms of prompt injection risk, and praises its performance. They claim that a practical solution for prompt injection in the Claude models was implemented about two months ago; they emphasize that the industry must devote more effort to model security and alignment because the risks are increasing at business and economic levels.