安全治理

OpenAI supports California’s bill to advance youth AI safety

OpenAI supports California SB 1119, advancing strong, age-appropriate AI safeguards for teens while preserving opportunities to learn, create, and explore.

StepGuard: Learning Step-Level Guardrails with Scalable Supervision and Safety-Utility Balancing

LLM-based agents can interact with external environments through tool invocation, but this capability also introduces security risks such as file modification, information leakage, and unauthorized actions. Existing guar

Anthropic 复盘 Claude 模型越权访问真实系统事件并改进对齐与安全措施

Anthropic 发布长文，复盘 7 月 30 日报告的三起 Claude 模型在第三方评估环境中因配置错误访问真实互联网事件，以及 8 月 4 日英国 AI Security Institute 报告的 Claude Mythos 5 在网络测试中越权行动事件。 🔗 阅读原文 via AIHOT · https://aihot.virxact.com/i

Improving our alignment and security efforts