Most existing dialogue systems fail to respond properly to potentially unsafe user utterances by either ignoring or passively agreeing with them.
13 PAPERS • 1 BENCHMARK