ORISA Accepted to IEEE SP 2026

Our paper, Recovering and Rehosting Mobile Local LLM Conversations and Contexts via Memory Forensics has been accepted to the 47th IEEE S&P Symposium.

Here is a synopsis of it, more to come in a future blog post with the link to the presentation and paper:

The rise of local large language models (LLMs) on mobile devices has significantly enhanced user privacy, but it also creates a challenge for forensic investigators when suspects use these apps to plan crimes and subsequently delete their chat histories. To address this critical gap, our new paper introduces RELIC, a pioneering memory forensics framework that recovers deleted conversation contexts by extracting batch tokens and key-value (KV) caches directly from the app’s process memory. By reconstructing these artifacts into an app-agnostic live session, RELIC allows investigators to effectively “rehost” the forgotten context and interrogate the model to reveal a suspect’s prior interactions and hidden motives. Thank you CyFI Lab!!