A team of researchers argues that large language models cannot be made fully secure against hacks because of a fundamental flaw in the way they operate. The claim was presented in a paper at the International Conference on Machine Learning (ICML), a leading AI conference, this month.
Why it matters
If the argument holds, it suggests that certain vulnerabilities in LLMs may be inherent rather than fixable through incremental patches. The researchers say this has significant implications for the safety of the technology.
Who should care
AI researchers, developers working with LLMs, and organizations relying on these models for security-sensitive applications may want to follow this work and its reception at ICML.