Detecting Verbatim LLM Copy-Paste in Homework

This collection highlights emerging security implications of advanced generative and agentic AI plus related defensive tools and governance work. Key papers: SteganoPrompt — an input-side, invisible Unicode-tag watermark that survives common copy/paste channels to detect verbatim LLM copy-paste; interaction-layer antidistillation watermarks — behavioral markers (follow-ups/restatements) that transfer into distilled models and resist some paraphrasing; “Infinite Impostor” and agentic-AI analyses — argue agentic systems collapse the fidelity-vs-scale tradeoff enabling high‑fidelity, mass-targets