美国指南已生效模型提供方

模型对齐偏差调查框架及报告

Framework for Tracking, Investigating, and Disclosing Instances of Model Misalignment

OpenAI·发布 2026-09-16·管理员核对于 2026-09-18

OpenAI发布新的框架以追踪、调查和公开披露模型对齐偏差，并公布了六份关于模型在训练和评估中表现偏差的报告。该框架设定了公开披露的标准和时限，优先处理揭示新对齐机制、已知行为重大变化或挑战安全假设的发现。这是OpenAI持续公开报告的起点。

原文 ↗提醒我美国 · 全部

对企业意味着什么

• Obligations: AI companies must follow OpenAI's framework for disclosing model misalignment issues to the public, especially for significant or safety-related findings.
• Opportunities: Companies can learn from OpenAI's public reports and framework to improve their own model alignment and safety practices.
• Deadlines: There are defined timelines for OpenAI to investigate and disclose findings, providing a clearer expectation for stakeholders.