AI RESEARCH
PragLocker: Protecting Agent Intellectual Property in Untrusted Deployments via Non-Portable Prompts
arXiv CS.AI
•
ArXi:2605.05974v1 Announce Type: cross LLM agents rely on prompts to implement task-specific capabilities based on foundation LLMs, making agent prompts valuable intellectual property. However, in untrusted deployments, adversaries can copy and reuse these prompts with other