Search papers, labs, and topics across Lattice.
1
0
3
LLM cybersecurity refusal policies can be made more consistent and less restrictive by explicitly modeling the trade-off between offensive risk and defensive benefit, rather than relying on intent or offensive classification.
