PRISM: Transfer and Limits of Activation Interventions in Tool-Calling Language Models
Recent literature shows that unsafe tool calls can often accompany textual refusals in language models. To investigate this hidden vulnerability, we introduce PRISM, a framework for evaluating the transfer of activation interventions across precision, languages, and scales. Our findings demonstrate that leveraging lowe...
Giovana Kerche Bonás, T. Almeida, Marcelo Henrique da Silva Reis et al.
· 0 citations