Towards evaluations-based safety cases for AI scheming

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Balesni, Mikita, Hobbhahn, Marius, Lindner, David, Meinke, Alexander, Korbak, Tomek, Clymer, Joshua, Shlegeris, Buck, Scheurer, Jérémy, Stix, Charlotte, Shah, Rusheb, Goldowsky-Dill, Nicholas, Braun, Dan, Chughtai, Bilal, Evans, Owain, Kokotajlo, Daniel, Bushnaq, Lucius
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916471568859136
author Balesni, Mikita
Hobbhahn, Marius
Lindner, David
Meinke, Alexander
Korbak, Tomek
Clymer, Joshua
Shlegeris, Buck
Scheurer, Jérémy
Stix, Charlotte
Shah, Rusheb
Goldowsky-Dill, Nicholas
Braun, Dan
Chughtai, Bilal
Evans, Owain
Kokotajlo, Daniel
Bushnaq, Lucius
author_facet Balesni, Mikita
Hobbhahn, Marius
Lindner, David
Meinke, Alexander
Korbak, Tomek
Clymer, Joshua
Shlegeris, Buck
Scheurer, Jérémy
Stix, Charlotte
Shah, Rusheb
Goldowsky-Dill, Nicholas
Braun, Dan
Chughtai, Bilal
Evans, Owain
Kokotajlo, Daniel
Bushnaq, Lucius
contents We sketch how developers of frontier AI systems could construct a structured rationale -- a 'safety case' -- that an AI system is unlikely to cause catastrophic outcomes through scheming. Scheming is a potential threat model where AI systems could pursue misaligned goals covertly, hiding their true capabilities and objectives. In this report, we propose three arguments that safety cases could use in relation to scheming. For each argument we sketch how evidence could be gathered from empirical evaluations, and what assumptions would need to be met to provide strong assurance. First, developers of frontier AI systems could argue that AI systems are not capable of scheming (Scheming Inability). Second, one could argue that AI systems are not capable of posing harm through scheming (Harm Inability). Third, one could argue that control measures around the AI systems would prevent unacceptable outcomes even if the AI systems intentionally attempted to subvert them (Harm Control). Additionally, we discuss how safety cases might be supported by evidence that an AI system is reasonably aligned with its developers (Alignment). Finally, we point out that many of the assumptions required to make these safety arguments have not been confidently satisfied to date and require making progress on multiple open research problems.
format Preprint
id arxiv_https___arxiv_org_abs_2411_03336
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Towards evaluations-based safety cases for AI scheming
Balesni, Mikita
Hobbhahn, Marius
Lindner, David
Meinke, Alexander
Korbak, Tomek
Clymer, Joshua
Shlegeris, Buck
Scheurer, Jérémy
Stix, Charlotte
Shah, Rusheb
Goldowsky-Dill, Nicholas
Braun, Dan
Chughtai, Bilal
Evans, Owain
Kokotajlo, Daniel
Bushnaq, Lucius
Cryptography and Security
Artificial Intelligence
We sketch how developers of frontier AI systems could construct a structured rationale -- a 'safety case' -- that an AI system is unlikely to cause catastrophic outcomes through scheming. Scheming is a potential threat model where AI systems could pursue misaligned goals covertly, hiding their true capabilities and objectives. In this report, we propose three arguments that safety cases could use in relation to scheming. For each argument we sketch how evidence could be gathered from empirical evaluations, and what assumptions would need to be met to provide strong assurance. First, developers of frontier AI systems could argue that AI systems are not capable of scheming (Scheming Inability). Second, one could argue that AI systems are not capable of posing harm through scheming (Harm Inability). Third, one could argue that control measures around the AI systems would prevent unacceptable outcomes even if the AI systems intentionally attempted to subvert them (Harm Control). Additionally, we discuss how safety cases might be supported by evidence that an AI system is reasonably aligned with its developers (Alignment). Finally, we point out that many of the assumptions required to make these safety arguments have not been confidently satisfied to date and require making progress on multiple open research problems.
title Towards evaluations-based safety cases for AI scheming
topic Cryptography and Security
Artificial Intelligence
url https://arxiv.org/abs/2411.03336