Search papers, labs, and topics across Lattice.
5
0
6
This work shows that a weaker, unaligned model can split a harmful task into benign-looking subproblems, consult a stronger aligned model independently on each, and combine the answers locally to laundering attack capability laundering.
SpecGuard is introduced, an inference-time backdoor detector that repurposes speculative decoding at zero added model-computation cost and doubles as a free, always-on signal for detecting backdoored LLM behavior.
Stripped of safety, language models can be tricked into generating misleading yet confident responses, with up to 90% of outputs being decoys under attack.
One in twenty papers from top-tier conferences may contain multiple hallucinated citations, raising serious concerns about the integrity of the academic record.
AdvGRPO enables robust attacker-defender co-training that significantly improves defender performance on safety benchmarks while generating effective attacks.
