Search papers, labs, and topics across Lattice.
This paper investigates the limitations of current safeguards in large language models (LLMs) that assess whether to provide answers without understanding their downstream use, highlighting the vulnerability of these systems to dual-use scenarios. The authors establish a safety trilemma, demonstrating that Useful Capability, Reliable Safety, and Open Access cannot coexist when evidence is easily copyable, which allows attackers to exploit benign requests. By introducing a trusted credential system that incorporates hard-to-copy information, the study proposes a method to enhance safety while maintaining utility, supported by empirical evaluations of dual-use scenarios and adaptive attacks.
A safety trilemma reveals that LLMs cannot balance utility, safety, and access when evidence is copyable, exposing a critical vulnerability in current safeguards.
Large language model safeguards decide whether to answer before seeing how an answer will be used. This creates a basic problem for dual-use tasks: the same answer can help an authorized professional or an attacker, while an attacker can imitate a benign request and interaction history. We separate the capability released by the model from the evidence available about downstream use. When that evidence is copyable, we derive the exact worst-case floor on attacker assistance while preserving useful answers. The result yields a safety trilemma: Useful Capability, Reliable Safety, and Open Access cannot coexist. We then show how a trusted credential can complement existing safeguards by adding hard-to-copy information that predicts actual downstream use, and identify the stronger condition needed to eliminate the floor. Evidence from dual-use evaluations, adaptive attacks, and deployed trusted-access programs supports the practical relevance of these conditions.
