Scaffolding and fine-tuning let LLM agents treat threats to surrogate goals like preventing money burning equivalently to direct threats, outperforming prompting with fewer capability side effects.
Threat text\
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.AI 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Implementing surrogate goals for safer bargaining in LLM-based agents
Scaffolding and fine-tuning let LLM agents treat threats to surrogate goals like preventing money burning equivalently to direct threats, outperforming prompting with fewer capability side effects.