8%) left revision traces.
LLM assistants can accidentally disclose withdrawn secrets in revision comments
A study finds that about 13% of deliverables from revised requests allow recovery of the withdrawn item from the model's explanation of the edit.
Research Lab
Yage Zhang · Yukun Jiang · Yang Zhang
CISPA Helmholtz Center for Information Security
Research Digest··2 min read
Zhang et al.
Why this paper
From CISPA Helmholtz Center for Information Security
In one line
Revision traces in LLM deliverables can disclose previously removed sensitive information, enabling third-party recovery.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ✓Reports numbers on named benchmarks (2 benchmarks)
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§