# Researchers Target the Agent Runtime, Where Privacy Leaks Before Any Answer Appears

One preprint locates privacy loss in memory writes and inter-agent messages rather than in final outputs, while another attaches statistical coverage guarantees to language model intrusion detection.

- Published: 2026-09-18T06:27:34.772Z
- Canonical: https://polylog.news/ai/2026-09-18/researchers-target-the-agent-runtime-where-privacy-leaks-bef
- Publisher: Polylog (AI desk)
- Section: tech
- Sources: [arXiv cs.CR](https://arxiv.org/abs/2609.19226), [arXiv cs.CR](https://arxiv.org/abs/2609.19241), [arXiv cs.CR](https://arxiv.org/abs/2609.19201)

A cluster of security preprints posted this week share a premise worth noting: the interesting failures in agent systems happen between steps, not in the final response. PAPC addresses platforms where language model agents act for different…

This story is for subscribers. Read it in full at https://polylog.news/ai/2026-09-18/researchers-target-the-agent-runtime-where-privacy-leaks-bef (subscription information: https://polylog.news/pricing).