Unified Threat Detection and Mitigation Framework (UTDMF): Combating Prompt Injection, Deception, and Bias in Enterprise-Scale Transformers
On the map: LLM safety and security → Prompt injection and agent security
Abstract (arXiv)
The rapid adoption of large language models (LLMs) in enterprise systems exposes vulnerabilities to prompt injection attacks, strategic deception, and biased outputs, threatening security, trust, and fairness. Extending our adversarial activation patching framework (arXiv:2507.09406), which induced deception in toy networks at a 23.9% rate, we introduce the Unified Threat Detection and Mitigation Framework (UTDMF), a scalable, real-time pipeline for enterprise-grade models like Llama-3.1 (405B), GPT-4o, and Claude-3.5. Through 700+ experiments per model, UTDMF achieves: (1) 92% detection accuracy for prompt injection (e.g., jailbreaking); (2) 65% reduction in deceptive outputs via enhanced patching; and (3) 78% improvement in fairness metrics (e.g., demographic bias). Novel contributions include a generalized patching algorithm for multi-threat detection, three groundbreaking hypotheses on threat interactions (e.g., threat chaining in enterprise workflows), and a deployment-ready toolkit with APIs for enterprise integration.
Key facts. Cited by 1 paper in the map (data through 7 Oct 2026). Placed in Prompt injection and agent security, one of 12 directions in LLM safety and security. Builds on 23 of its 51 references that are in the map. First posted on arXiv 6 Oct 2025.
Builds on
23 of 51 references.
Closest ancestors · 4
Same direction, newest first
- 2025SafeArena: Evaluating the Safety of Autonomous Web Agents6 Mar
- 2024Attention Tracker: Detecting Prompt Injection Attacks in LLMs1 Nov
- 2024Goal-guided Generative Prompt Injection Attack on Large Language Models6 Apr
- 2024Automatic and Universal Prompt Injection Attacks against Large Language Models7 Mar
Other references · 19
- 2024Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Trainingcited by 478
- 2023AI Deception: A Survey of Examples, Risks, and Potential Solutionscited by 185
- 2018The Malicious Use of Artificial Intelligence: Forecasting, Prevention, and Mitigationcited by 159
- 2023Large Language Models can Strategically Deceive their Users when Put Under Pressurecited by 132
- 2023Generative Language Models Exhibit Social Identity Biasescited by 107
- 2024Large Language Models for Cyber Security: A Systematic Literature Reviewcited by 65
- 2024Fairness in Large Language Models: A Taxonomic Surveycited by 57
- 2025LLMs in Software Security: A Survey of Vulnerability Detection Techniques and Insightscited by 44
- 2025LLMs for Explainable AI: A Comprehensive Surveycited by 42
- 2025The Traitors: Deception and Trust in Multi-Agent Language Model Simulationscited by 30
- 2025Frontier AI's Impact on the Cybersecurity Landscapecited by 21
- 2025When Thinking LLMs Lie: Unveiling the Strategic Deception in Representations of Reasoning Modelscited by 16
7 more
- 2025Deception in LLMs: Self-Preservation and Autonomous Goals in Large Language Modelscited by 15
- 2025A Frontier AI Risk Management Framework: Bridging the Gap Between Current AI Practices and Established Risk Managementcited by 8
- 2025Emerging Practices in Frontier AI Safety Frameworkscited by 6
- 2025BEATS: Bias Evaluation and Assessment Test Suite for Large Language Modelscited by 5
- 2025eX-NIDS: A Framework for Explainable Network Intrusion Detection Leveraging Large Language Modelscited by 5
- 2025Adversarial Activation Patching: A Framework for Detecting and Mitigating Emergent Deception in Safety-Aligned Transformerscited by 2
- 2025Biases in Edge Language Models: Detection, Analysis, and Mitigationcited by 1
Built on by
Citing papers per year: 2026: 1.
1 paper.