Anthropic PBC 计划让企业客户在使用其最强大的人工智慧模型时，能够对自身资料拥有更大的控制权。此举与该公司先前为防范潜在网路攻击而推出的资料留存政策形成鲜明对比。新的安全系统预计于今年稍后推出，仍要求企业客户保留资料三十天，但客户可选择将资料存放在自己的云端基础设施上，而非 Anthropic 的伺服器中。（关键数字：30）

Anthropic 于六月宣布，将针对其更强大的 Mythos 和 Fable 模型以及未来的前沿产品，留存所有客户资料三十天，以协助侦测与防止新型网路攻击。该公司当时表示不会将这些资料用于训练 AI，但也承认此政策可能不受客户欢迎，并对其商业成功构成实际风险，尤其是在竞争对手未采取相同措施的情况下。（关键数字：30）

Anthropic 在用户开始表达批评之前便已著手开发这套新系统，并已与超过一百家处于高度监管行业的客户进行协调。与此同时，其长期竞争对手 OpenAI 也于周三宣布开始测试名为「私密安全处理」的方法，以避免留存用户资料，并计划于九月更广泛地发布该功能。（关键数字：100）

Anthropic PBC plans to give enterprise customers greater control over their data when using its most capable AI models, reversing an earlier data retention policy designed to mitigate potential cyberattacks. A new safety system expected later this year will still require a 30-day data retention period, but customers will have the option to store that data on their own cloud infrastructure instead of Anthropic's servers.

In June, Anthropic announced it would retain all customer data for its powerful Mythos and Fable models, as well as future frontier offerings, for 30 days to help detect and prevent novel cyberattacks. While the company pledged not to use the data for AI training, it acknowledged in a recent report that the policy risked being unpopular with customers accustomed to zero retention and could harm its competitive standing if rivals did not adopt similar measures.

Anthropic had already begun developing the new system before users voiced criticism, and it has been coordinating with over 100 customers in highly regulated industries to refine the option. Meanwhile, rival OpenAI announced it started testing its own approach called private safety processing, partnering with companies including Databricks and Microsoft, with a broader rollout planned for September.