On August 20th, OpenAI announced that it will enhance security handling processes for paid users and initiate testing of the 'Private Security Handling' technology. This technology improves security protection by identifying risk patterns across multiple interactions, rather than analyzing just a single prompt and response. OpenAI pointed out that as AI undertakes more long-term and autonomous tasks, risks are often hidden in combinations of multiple interactions. The new process aims to defend against attacks from malicious users and out-of-control AI agents, while ensuring compatibility with 'zero data retention,' meaning customer content will still not be accessible to OpenAI. OpenAI plans to officially launch this feature in September and release a technical white paper, currently collaborating with early customers such as Microsoft and Databricks for testing.