{"id":{"repo_id":"mit","oai_identifier":"oai:dspace.mit.edu:1721.1/159116"},"canonical_url":"https://search.dev.ndltd.org/etd/mit/oai:dspace.mit.edu:1721.1/159116","repository":{"repo_id":"mit","name":"MIT","base_url":"https://dspace.mit.edu/oai/request"},"display":{"title":"Exploring Fine-Tuning Techniques for Removing Tamper-Resistant Safeguards for Open-Weight LLMs","abstract":"Open-source models present significant opportunities and risks, especially in dual-use scenarios where they can be repurposed for malicious tasks via adversarial fine-tuning. In this paper, we evaluate the effectiveness of Tampering Attack Resistance (TAR), a safeguard designed to protect against such adversarial attacks, by exploring its resilience to full-parameter and parameter-efficient fine-tuning. Our experiments reveal that while TAR enhances tamper resistance compared to models without safeguards, it remains susceptible to variability. Specifically, we observe inconsistencies where the same adversarial attack can succeed under some initializations and fail under others. This is a critical security risk as even a single instance of failure can lead to models being exploited for harmful purposes. These findings highlight the limitations of current tamper-resistant safeguards and emphasize the need for more robust safeguards to ensure the safe and ethical deployment of open-source models.","abstract_html":"Open-source models present significant opportunities and risks, especially in dual-use scenarios where they can be repurposed for malicious tasks via adversarial fine-tuning. In this paper, we evaluate the effectiveness of Tampering Attack Resistance (TAR), a safeguard designed to protect against such adversarial attacks, by exploring its resilience to full-parameter and parameter-efficient fine-tuning. Our experiments reveal that while TAR enhances tamper resistance compared to models without safeguards, it remains susceptible to variability. Specifically, we observe inconsistencies where the same adversarial attack can succeed under some initializations and fail under others. This is a critical security risk as even a single instance of failure can lead to models being exploited for harmful purposes. These findings highlight the limitations of current tamper-resistant safeguards and emphasize the need for more robust safeguards to ensure the safe and ethical deployment of open-source models.","abstract_has_math":false,"creators":["Zhang, Sarah"],"institution":"Massachusetts Institute of Technology","degree_name":"Master","degree_level":null,"degree_discipline":null,"degree_department":"Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science","school":null,"contributors":[],"advisors":["Kim, Yoon"],"committee_chairs":[],"committee_members":[],"year":2025,"date_issued":"2025-02","date_published":"2025-02","updated_at":"2026-07-22T22:21:56Z","subjects":[],"languages":[],"rights":["In Copyright - Educational Use Permitted","Copyright retained by author(s)"],"rights_urls":["https://rightsstatements.org/page/InC-EDU/1.0/"],"identifier_entries":[]},"links":{"outbound_url":"https://hdl.handle.net/1721.1/159116","outbound_label":"Handle","outbound_source":"dc:identifier.uri"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor.advisor","label":"Advisor","values":["Kim, Yoon"]},{"key":"dc:contributor.department","label":"Department","values":["Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science"]},{"key":"dc:creator","label":"Author","values":["Zhang, Sarah"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.accessioned","label":"Dc Date Accessioned","values":["2025-04-14T14:06:35Z"]},{"key":"dc:date.available","label":"Dc Date Available","values":["2025-04-14T14:06:35Z"]},{"key":"dc:date.issued","label":"Date","values":["2025-02"]},{"key":"dc:publisher","label":"Institution","values":["Massachusetts Institute of Technology"]},{"key":"dc:type","label":"Dc Type","values":["Thesis"]},{"key":"thesis:degree_name","label":"Degree Name","values":["Master","Master of Engineering in Electrical Engineering and Computer Science"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:rights","label":"Dc Rights","values":["In Copyright - Educational Use Permitted","Copyright retained by author(s)"]},{"key":"dc:rights.uri","label":"Rights URI","values":["https://rightsstatements.org/page/InC-EDU/1.0/"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier.uri","label":"Identifier URI","values":["https://hdl.handle.net/1721.1/159116"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["Open-source models present significant opportunities and risks, especially in dual-use scenarios where they can be repurposed for malicious tasks via adversarial fine-tuning. In this paper, we evaluate the effectiveness of Tampering Attack Resistance (TAR), a safeguard designed to protect against such adversarial attacks, by exploring its resilience to full-parameter and parameter-efficient fine-tuning. Our experiments reveal that while TAR enhances tamper resistance compared to models without safeguards, it remains susceptible to variability. Specifically, we observe inconsistencies where the same adversarial attack can succeed under some initializations and fail under others. This is a critical security risk as even a single instance of failure can lead to models being exploited for harmful purposes. These findings highlight the limitations of current tamper-resistant safeguards and emphasize the need for more robust safeguards to ensure the safe and ethical deployment of open-source models."]},{"key":"dc:description.degree","label":"Dc Description Degree","values":["M.Eng."]},{"key":"dc:title","label":"Title","values":["Exploring Fine-Tuning Techniques for Removing Tamper-Resistant Safeguards for Open-Weight LLMs"]}]}],"canonical_facts":{"dc:contributor.advisor":["Kim, Yoon"],"dc:contributor.department":["Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science"],"dc:creator":["Zhang, Sarah"],"dc:date.accessioned":["2025-04-14T14:06:35Z"],"dc:date.available":["2025-04-14T14:06:35Z"],"dc:date.issued":["2025-02"],"dc:description.abstract":["Open-source models present significant opportunities and risks, especially in dual-use scenarios where they can be repurposed for malicious tasks via adversarial fine-tuning. In this paper, we evaluate the effectiveness of Tampering Attack Resistance (TAR), a safeguard designed to protect against such adversarial attacks, by exploring its resilience to full-parameter and parameter-efficient fine-tuning. Our experiments reveal that while TAR enhances tamper resistance compared to models without safeguards, it remains susceptible to variability. Specifically, we observe inconsistencies where the same adversarial attack can succeed under some initializations and fail under others. This is a critical security risk as even a single instance of failure can lead to models being exploited for harmful purposes. These findings highlight the limitations of current tamper-resistant safeguards and emphasize the need for more robust safeguards to ensure the safe and ethical deployment of open-source models."],"dc:description.degree":["M.Eng."],"dc:identifier.uri":["https://hdl.handle.net/1721.1/159116"],"dc:publisher":["Massachusetts Institute of Technology"],"dc:rights":["In Copyright - Educational Use Permitted","Copyright retained by author(s)"],"dc:rights.uri":["https://rightsstatements.org/page/InC-EDU/1.0/"],"dc:title":["Exploring Fine-Tuning Techniques for Removing Tamper-Resistant Safeguards for Open-Weight LLMs"],"dc:type":["Thesis"],"thesis:degree_name":["Master","Master of Engineering in Electrical Engineering and Computer Science"]},"updated_at":"2026-07-22T22:21:56Z"}