Library of Congress Offers AI Legal Guidance to Researchers

**Executive summary:** The Library of Congress clarified that some offensive AI security testing activities (such as prompt injection and bypassing rate limits) do not violate the DMCA, but rejected a broader fair-use exemption for AI security researchers, leaving gaps in legal safe harbor and potential liability under DMCA Section 1201 and the CFAA; the report captures community reactions and emphasizes that while red teaming is valuable, lasting AI trustworthiness requires design- and data-focused improvements rather than reliance on pen-testing alone.