Implementing surrogate goals for safer bargaining in LLM-based agents

Researchers developed methods to implement 'surrogate goals' in LLM-based agents to reduce bargaining risks by deflecting threats away from what principals care about. The study tested four approaches (prompting, fine-tuning, scaffolding) and found that scaffolding and fine-tuning methods outperformed simple prompting for implementing desired threat response behaviors.

• →Surrogate goals are designed to redirect threats in AI agent bargaining away from what principals value most.
• →Four implementation methods were tested: prompting, fine-tuning, and two scaffolding approaches.
• →Fine-tuning and scaffolding methods more precisely implemented desired threat response behaviors than simple prompting.
• →Scaffolding-based methods showed the best performance with fewer negative side effects on other capabilities.
• →The research addresses AI safety concerns in multi-agent bargaining scenarios involving LLMs.