ConMem Algorithm Resolves LLM Memory Misalignment via Counterfactual Probing

The ConMem algorithm addresses memory misalignment in long-horizon agents by using prospective counterfactual causal inference. Achieving 75.16% accuracy, it prevents agents from inappropriately retaining over-generalized or privacy-sensitive information during multi-session tasks.