OpenAPA security method uses military-grade data isolation to block LLM prompt injection

OpenAPA presents a new approach to prevent prompt injection attacks in LLM-based agents. The method applies military-style data classification, where each resource and destination receives a security label. An agent session inherits the highest classification level of any data it accesses. Any attempt to transfer information to a lower-classification destination is then automatically blocked, independent of the agent's instructions or manipulation attempts.