SuperLocalMemory: Privacy-Preserving Multi-Agent Memory with Bayesian Trust Defense Against Memory Poisoning

Summary

When several agents write to one memory store, a single compromised agent can poison what every other agent later recalls. This paper models writer trust as a Bayesian posterior updated by downstream outcomes, so a source that keeps producing facts which fail verification loses influence over retrieval — without any of the data leaving the machine.

Our summary of the paper. The full abstract is on arXiv. Preprint, not peer reviewed.

Authors

• Varun Pratap Bhardwaj

Cite this paper

@misc{bhardwaj2026slmprivacy,
  title        = {SuperLocalMemory: Privacy-Preserving Multi-Agent Memory with Bayesian Trust Defense Against Memory Poisoning},
  author       = {Varun Pratap Bhardwaj},
  year         = {2026},
  eprint       = {2603.02240},
  archivePrefix= {arXiv},
  primaryClass = {cs.AI},
  url          = {https://arxiv.org/abs/2603.02240}
}