Researchers withdraw a paper on LLM moderation vulnerabilities after identifying experimental errors.