AI guard model packs safety reasoning into compressed states, cutting costs

Original: LatentGuard: Efficient and Inspectable Latent Reasoning for LLM Safeguards

Source: arxiv.org ↗