src_fb50be84b835134c75
- Source-provided label
- Not reported
- Evidence class
- mentioned in report
- Confidence
- Not reported
- Observed
- Oct 07, 2026 04:00 UTC
- Retrieved
- Oct 08, 2026 03:43 UTC
Harmful SFT Leaves a Continuous Trace in LLM Checkpoint Updates
arXiv:2610.07518v1 Announce Type: cross Abstract: Safety auditing of post-trained large language models typically relies on model behavior, requiring model execution and depending on the coverage of available evaluations. This work asks a different question: Do the target behaviors optimized during supervised fine-tuning (SFT) leave readable evidence directly in checkpoint updates? We find that ha
Reference details
- Source category
- Research & news
- Report title
Harmful SFT Leaves a Continuous Trace in LLM Checkpoint Updates- First reported
- Oct 07, 2026 04:00 UTC
- Last reported
- Oct 07, 2026 04:00 UTC
- Catalog updated
- Oct 08, 2026 03:43 UTC
- Record ID
745772f9efe3ad2722ba8438badfd977cd1839c0a98cda38244edd1e032e950d
Published indicators
Network indicators are visibly defanged. This page does not make them clickable or send them to another service.
No validated file hashes or network indicators are stored for this record.
Source observations
This record contains metadata and attribution only. It does not host, download, unpack, render, or execute malware samples.
Feed collection status
Sources represented on this page. Publication and observation dates describe the content; successful batch dates describe local collection. A recent batch does not certify complete coverage or current provider availability.
1 sources: 0 stale, 0 needing attention, 0 paused, 1 unverified.
Review source update dates and status
- src_fb50be84b835134c75Collection status unverifiedLast successful batch: Unverified
Stale means older than twice the recorded collection interval, with a minimum of one hour; without an interval, the threshold is 24 hours. Cached content remains available while a feed is stale, paused, or unverified.