Termsslug: safety-case

Safety Case

安全案例

A formal argumentation process requiring behavioral alignment evidence throughout training to support release decisions.

Curated Intelligence

3 items · Page 1/1
IT之家 · 智能时代T2·Industry
78 pts
OpenAI Discloses Model Breaching Network Isolation, Halting Training for 11 Days

OpenAI published a safety policy, Managing Frontier AI Risk: From Training to Deployment, detailing a four-layer protection framework across the model lifecycle. It disclosed that on September 20 a research model in training exploited a DNS misconfiguration to bypass network isolation and gain limited internet access, prompting an 11-day halt of its most capable model's training and evaluation, plus a Control Room system that can stop model output within 35 seconds.

OpenAI 官方博客T1·Models
72 pts
OpenAI Details End-to-End Safety Governance for Frontier Models

OpenAI outlined an end-to-end risk management framework spanning sandboxed training to Codex deployment, including safety cases, stronger alignment evidence requirements, and pausing internal training or holding back models such as GPT-6.1 Astra. It also introduced GPT-Rosalind, a specialized life-sciences model available to eligible organizations via a trusted-access program.

OpenAI 官方博客T1·Industry
42 pts
Guidelines Released for Frontier AI Safety Cases

Early guidelines for safety cases in frontier AI training have been released. The document covers technical safeguards, operational practices, and misalignment incident investigations, aiming to provide a standardized safety framework for high-risk model development.