{"id":{"repo_id":"msu","oai_identifier":"oai:d.lib.msu.edu:etd_53224"},"canonical_url":"https://search.ndltd.org/etd/msu/oai:d.lib.msu.edu:etd_53224","repository":{"repo_id":"msu","name":"Michigan State University","base_url":"https://d.lib.msu.edu/oai"},"display":{"title":"The Evolution of AI Security : From Models to Agents","abstract":"Thesis (Ph.D.)--Michigan State University. Computer Science - Doctor of Philosophy, 2026","abstract_html":"Thesis (Ph.D.)--Michigan State University. Computer Science - Doctor of Philosophy, 2026","abstract_has_math":false,"creators":["He, Pengfei"],"institution":null,"degree_name":null,"degree_level":null,"degree_discipline":null,"degree_department":null,"school":null,"contributors":["Tang, Jiliang","Cui, Yuehua","Yamada, Makoto","Liu, Sijia"],"advisors":[],"committee_chairs":[],"committee_members":[],"year":2026,"date_issued":"2026","date_published":"2026","updated_at":"2026-11-18T18:59:44Z","subjects":["Computer science","Statistics"],"languages":["eng"],"rights":["Attribution 4.0 International"],"rights_urls":[],"identifier_entries":[{"key":"dc:identifier","label":"Identifier","values":["etd:53224","local:He_grad.msu_0128D_22267"],"render_values":[{"text":"etd:53224","href":null,"code":true},{"text":"local:He_grad.msu_0128D_22267","href":null,"code":true}]}],"person_identifier_entries":[]},"links":{"outbound_url":"https://doi.org/doi:10.25335/4bet-dd31","outbound_label":"DOI","outbound_source":"dc:identifier","record_url":"https://doi.org/doi:10.25335/4bet-dd31","record_label":"Repository record","record_kind":"doi","record_source":"dc:identifier","download_url":null,"download_label":null,"download_source":null,"access_status":null,"embargo_until":null,"access_note":null},"source_record":{"url":"https://d.lib.msu.edu/oai?verb=GetRecord&metadataPrefix=oai_dc&identifier=oai%3Ad.lib.msu.edu%3Aetd_53224","prefix":"oai_dc"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor","label":"Contributor","values":["Tang, Jiliang","Cui, Yuehua","Yamada, Makoto","Liu, Sijia"]},{"key":"dc:creator","label":"Author","values":["He, Pengfei"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date","label":"Dc Date","values":["2026"]},{"key":"dc:relation","label":"Dc Relation","values":["Electronic Theses & Dissertations"]},{"key":"dc:type","label":"Dc Type","values":["text","Electronic dissertations"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Computer science","Statistics"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language","label":"Dc Language","values":["eng"]},{"key":"dc:rights","label":"Dc Rights","values":["Attribution 4.0 International"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier","label":"Identifier","values":["etd:53224","local:He_grad.msu_0128D_22267","https://doi.org/doi:10.25335/4bet-dd31"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description","label":"Description","values":["Thesis (Ph.D.)--Michigan State University. Computer Science - Doctor of Philosophy, 2026","Deep learning models, particularly large language models (LLMs), are increasingly deployed in high-stakes applications, making their security and reliability critical concerns. As AI systems evolve from traditional predictive models to LLMs and autonomous multi-agent systems, they introduce new vulnerabilities and attack surfaces that challenge existing security paradigms. This dissertation studies the evolution of AI security from models to agents and develops attacks, defenses, and theoretical foundations for building trustworthy intelligent systems.First, I investigate data poisoning attacks against deep learning models. I propose Sharpness-Aware Data Poisoning (SAPA), a transferable poisoning framework that remains effective under limited attacker knowledge by leveraging loss landscape sharpness. Second, I study security risks in LLMs and introduce the first data poisoning attack on in-context learning, demonstrating that carefully crafted examples can manipulate model behavior by influencing hidden representations. To better understand this emerging capability, I develop a statistical framework that characterizes how demonstrations affect in-context learning, revealing the roles of class separation, demonstration alignment, and label bias in shaping model predictions.Finally, I examine security challenges in LLM-based multi-agent systems. I develop the first attack targeting inter-agent communication, showing how compromised communication channels can manipulate collective behaviors. Building on these findings, I propose an attention-based trust management framework that enables agents to assess communication trustworthiness and defend against malicious collaborators.Together, these contributions advance the understanding and security of modern AI systems, spanning traditional deep learning models, large language models, and autonomous agent ecosystems. This work provides both practical security mechanisms and theoretical insights toward the development of secure, trustworthy, and resilient AI.","Description based on online resource. Title from PDF t.p. (Michigan State University Fedora Repository, viewed ).","Includes bibliographical references."]},{"key":"dc:format","label":"Dc Format","values":["application/pdf","1 online resource (218 p.) : ill."]},{"key":"dc:title","label":"Title","values":["The Evolution of AI Security : From Models to Agents"]}]}],"canonical_facts":{"dc:contributor":["Tang, Jiliang","Cui, Yuehua","Yamada, Makoto","Liu, Sijia"],"dc:creator":["He, Pengfei"],"dc:date":["2026"],"dc:description":["Thesis (Ph.D.)--Michigan State University. Computer Science - Doctor of Philosophy, 2026","Deep learning models, particularly large language models (LLMs), are increasingly deployed in high-stakes applications, making their security and reliability critical concerns. As AI systems evolve from traditional predictive models to LLMs and autonomous multi-agent systems, they introduce new vulnerabilities and attack surfaces that challenge existing security paradigms. This dissertation studies the evolution of AI security from models to agents and develops attacks, defenses, and theoretical foundations for building trustworthy intelligent systems.First, I investigate data poisoning attacks against deep learning models. I propose Sharpness-Aware Data Poisoning (SAPA), a transferable poisoning framework that remains effective under limited attacker knowledge by leveraging loss landscape sharpness. Second, I study security risks in LLMs and introduce the first data poisoning attack on in-context learning, demonstrating that carefully crafted examples can manipulate model behavior by influencing hidden representations. To better understand this emerging capability, I develop a statistical framework that characterizes how demonstrations affect in-context learning, revealing the roles of class separation, demonstration alignment, and label bias in shaping model predictions.Finally, I examine security challenges in LLM-based multi-agent systems. I develop the first attack targeting inter-agent communication, showing how compromised communication channels can manipulate collective behaviors. Building on these findings, I propose an attention-based trust management framework that enables agents to assess communication trustworthiness and defend against malicious collaborators.Together, these contributions advance the understanding and security of modern AI systems, spanning traditional deep learning models, large language models, and autonomous agent ecosystems. This work provides both practical security mechanisms and theoretical insights toward the development of secure, trustworthy, and resilient AI.","Description based on online resource. Title from PDF t.p. (Michigan State University Fedora Repository, viewed ).","Includes bibliographical references."],"dc:format":["application/pdf","1 online resource (218 p.) : ill."],"dc:identifier":["etd:53224","local:He_grad.msu_0128D_22267","https://doi.org/doi:10.25335/4bet-dd31"],"dc:language":["eng"],"dc:relation":["Electronic Theses & Dissertations"],"dc:rights":["Attribution 4.0 International"],"dc:subject":["Computer science","Statistics"],"dc:title":["The Evolution of AI Security : From Models to Agents"],"dc:type":["text","Electronic dissertations"]},"updated_at":"2026-11-18T18:59:44Z"}