Researchers map blind spots in AI safety attack benchmarks

Original: Talk is (Not) Cheap: A Taxonomy and Benchmark Coverage Audit for LLM Attacks

Source: arxiv.org ↗