AI models may hide their true reasoning, threatening security checks

Original: The Implications of Linguistic Illegibility for LLM Security

Source: arxiv.org ↗