phenomenon · first seen 6 Oct 2026
fine-tuning attack
Adversarial fine-tuning of an open-weight model to restore removed or suppressed harmful capabilities.
Mentions per week
PapersCodeDiscussion
Earliest sources
- Removing Information Content Does Not Certify Tamper Resistance in Open-Weight ModelsPapers · 6 Oct 2026 · Domenic Rosati, Alessa Carbo, Ali Dadsetan, Hong Huang, Matthew Young, Subhabrat…
Latest sources
- Removing Information Content Does Not Certify Tamper Resistance in Open-Weight ModelsPapers · 6 Oct 2026 · Domenic Rosati, Alessa Carbo, Ali Dadsetan, Hong Huang, Matthew Young, Subhabrat…