Georgia Tech Researchers Develop ORISA Tool to Recover Deleted AI Chats from Phone Memory

ORISA Brings Deleted AI Chats Back from Phone Memory

Georgia Tech researchers have developed a forensic tool called ORISA that can recover deleted conversations from local AI assistants running on Android phones and rebuild them into working sessions1

1

2

2

. The breakthrough reveals that hitting delete on an AI chat doesn't actually erase everything, as conversation data lingers in the phone's memory even after removal through the app interface.

Local AI assistants run directly on phones rather than sending conversations to cloud services. While these on-device AI apps operate, information from conversations remains in the phone's working memory even after deletion1

1

. ORISA searches this memory for fragments of deleted conversations and stitches them back together into live, working AI sessions that users can continue interacting with2

2

.

Recovery Capabilities Exceed Original Session Limits

In a controlled experiment, Georgia Tech researchers created a fictional AI assistant called Project Note Keeper using the ChatterUI Android app. They fed it 100 fictional facts about a project, then deleted the entire conversation1

1

. ORISA successfully recovered information from the deleted conversation and created a new working session. When researchers asked the restored assistant to list the project notes it had been given, it returned 26 facts1

1

. Notably, the original session could access only 14 facts because the rest had fallen outside its active memory window.

This demonstrates that ORISA can recover deleted conversations containing more information than what the original live session could actively manage at the time of deletion2

2

. The experiment proved researchers could recover information the AI had already stopped actively using, revealing data retention risks that extend beyond visible chat histories.

Extensive Testing Across Multiple Platforms

The researchers tested ORISA with 10 Android apps and three AI models using memory forensics techniques. Across 60 test configurations, the tool successfully recovered conversation information that remained in the phone's memory1

1

. The research was presented at the 2026 IEEE Symposium on Security and Privacy in a paper titled "Recovering and Rehosting Mobile Local LLM Conversations and Contexts via Memory Forensics"1

1

. The work was conducted in Georgia Tech's Cyber Forensics Innovation Laboratory by Ph.D. students Ken Xu, David Oygenblik, Runze Zhang and Mingxuan Yao, research scientist Muhammad Ibrahim, and associate professor Brendan Saltaformaggio.

Digital Investigations and Privacy Implications

According to lead author Ken Xu, ORISA "can help establish what information someone shared with an AI assistant and what assistance they received"1

1

. The tool has real forensic value for digital investigations, potentially helping investigators piece together conversations that may have legal or medical significance. Researchers note it could investigate cases where AI chatbot advice may have harmed patients, but the conversations were never shared with their doctors2

2

.

However, the same capability raises significant privacy concerns for people who use local LLMs on their devices. Deleting a conversation might not provide the clean break users expect when using local AI assistants2

2

. The findings challenge the assumption that on-device AI offers complete privacy simply because conversations aren't sent to the cloud.

Technical Limitations and Future Outlook

ORISA requires access to a copy of the phone's memory and storage to function, making it a specialized forensic tool rather than a consumer application2

2

. The researchers acknowledge that recovered information becomes harder to retrieve as the phone's working memory is overwritten by other activity1

1

. This means the window for successful recovery narrows as users continue using their devices after deletion.

The research puts a spotlight on how much local AI apps might be quietly holding onto behind the scenes, even after users believe they've erased sensitive conversations. As local LLMs become more prevalent on smartphones, understanding these data retention risks will become increasingly important for both users seeking privacy and investigators conducting digital forensics. Watch for potential policy discussions around mandatory memory-clearing standards for AI apps and clearer user disclosures about what deletion actually means in the context of on-device AI.