TRACE: Data-Free Text Reconstruction Attacks against LLM Unlearning

# TRACE: Data-Free Text Reconstruction Attacks against LLM Unlearning

https://gcdradar.vercel.app/incidents/trace-data-free-text-reconstruction-attacks-against-llm-unle · AIR-2026-1515 · closed 2026-11-09

- Lane: Research
- Severity: 1 (Research / proof of concept)
- Fix status: Fix status unknown
- First reported: 2026-10-26
- AI role: Attack on AI

## Summary

Researchers demonstrate a data-free text reconstruction attack that recovers residual traces left by approximate unlearning in large language model snapshots. The technique exposes potential privacy vulnerabilities when pre- and post-unlearning model snapshots are maintained for auditing.

## What to do

No action recorded; read the sources below and the vendor's advisory.

## Latest updates

- 2026-10-26: TRACE: Data-Free Text Reconstruction Attacks against Approximate Unlearning in LLMs (https://arxiv.org/abs/2610.31995)

## Sources

- 2026-10-26 arXiv: TRACE: Data-Free Text Reconstruction Attacks against Approximate Unlearning in LLMs https://arxiv.org/abs/2610.31995

Verify against the vendor's advisory before changing production. Data CC BY-SA 4.0, https://gcdradar.vercel.app.