LESSWRONG
LW

2743
Wikitags

AI Safety Cases

Edited by Rauno Arike last updated 19th Nov 2024

A safety case is a structured argument showing that a system is acceptably safe for a specific use in a specific environment. Safety cases typically include:

  • A description of the system's operational context
  • Identification of potential hazards and their consequences
  • A description of the risk controls that mitigate the hazards
  • An account of any residual risk
Subscribe
Discussion
1
Subscribe
Discussion
1
Posts tagged AI Safety Cases
31AXRP Episode 45 - Samuel Albanie on DeepMind’s AGI Safety Approach
Ω
DanielFilan
3mo
Ω
0
9Near- and medium-term AI Control Safety Cases
Ω
Martín Soto
10mo
Ω
0
145AI companies are unlikely to make high-assurance safety cases if timelines are short
Ω
ryan_greenblatt
9mo
Ω
5
95Anthropic: Three Sketches of ASL-4 Safety Case Components
Ω
Zach Stein-Perlman
1y
Ω
33
91New report: Safety Cases for AI
Ω
joshc
2y
Ω
14
60Toward Safety Cases For AI Scheming
Ω
Mikita Balesni, Marius Hobbhahn
1y
Ω
1
57A sketch of an AI control safety case
Ω
Tomek Korbak, joshc, Benjamin Hilton, Buck, Geoffrey Irving
9mo
Ω
0
49Notes on control evaluations for safety cases
Ω
ryan_greenblatt, Buck, Fabien Roger
2y
Ω
0
20The V&V method - A step towards safer AGI
Yoav Hollander
4mo
1
Add Posts