The Malicious Use of Artificial Intelligence: Forecasting, Prevention, and Mitigation

2018/02/20 by Miles Brundage, Shahar Avin, Brundage, Miles +49 · 6 voices · 19 citations
#cs.AI #cs.CR #cs.CY

paper · pdf · doi:10.48550/arxiv.1802.07228

Abstract

This report surveys the landscape of potential security threats from malicious uses of AI, and proposes ways to better forecast, prevent, and mitigate these threats. After analyzing the ways in which AI may influence the threat landscape in the digital, physical, and political domains, we make four high-level recommendations for AI researchers and other stakeholders. We also suggest several promising areas for further research that could expand the portfolio of defenses, or make attacks less effective or harder to execute. Finally, we discuss, but do not conclusively resolve, the long-term equilibrium of attackers and defenders.

Cited by

• Solipsistic Superintelligence is Unlikely to be Cooperative
• Super Suffixes: Bypassing Text Generation Alignment and Guard Models Simultaneously
• An Adaptive Multi-Layered Honeynet Architecture for Threat Behavior Analysis via Deep Learning
• Big Tech-Funded AI Papers Have Higher Citation Impact, Greater Insularity, and Larger Recency Bias
• AI & Human Co-Improvement for Safer Co-Superintelligence
• RoguePrompt: Dual-Layer Ciphering for Self-Reconstruction to Circumvent LLM Moderation
• Monte Carlo Expected Threat (MOCET) Scoring
• Hiding in the AI Traffic: Abusing MCP for LLM-Powered Agentic Red Teaming
• Three Lenses on the AI Revolution: Risk, Transformation, Continuity
• Unified Threat Detection and Mitigation Framework (UTDMF): Combating Prompt Injection, Deception, and Bias in Enterprise-Scale Transformers
• NeurIPS should lead scientific consensus on AI policy
• RoguePrompt: Dual-Layer Encoding for Self-Reconstruction to Circumvent LLM Moderation
• Journalists' Perceptions of Artificial Intelligence and Disinformation Risks
• Artificial Intelligence Training in Media: Addressing Technical and Ethical Challenges for Journalists and Media Professionals
• Ethics of Artificial Intelligence
• Passive Hack-Back Strategies for Cyber Attribution: Covert Vectors in Denied Environment
• STREAM (ChemBio): A Standard for Transparently Reporting Evaluations in AI Model Reports
• Securing Agentic AI: Threat Modeling and Risk Analysis for Network Monitoring Agentic AI System
• Dimensional Characterization and Pathway Modeling for Catastrophic AI Risks

Discussions

• Malicious Use of Artificial Intelligence: Forecasting, Prevention and Mitigation [hn, 4 points, 0 comments]
• The Malicious Use of Artificial Intelligence: Forecasting, Prevention,Mitigation [hn, 2 points, 0 comments]
• The Malicious Use of AI: Forecasting, Prevention, and Mitigation (2018) [pdf] [hn, 2 points, 1 comments]
• The Malicious Use of AI: Forecasting, Prevention, and Mitigation [hn, 1 points, 0 comments]
• The Malicious Use of AI: Forecasting, Prevention and Mitigation [hn, 1 points, 0 comments]
• 8/ Le papier auquel il a contribué de loin le plus cité (genre 10 fois + que la plupart de ses autres) remonte à 2018 et était une collaboration avec OpenAI sur les usages malveillants de l'IA, assez [bsky, 0 points, 1 comments]

Related

• Considerations Influencing Offense-Defense Dynamics From Artificial Intelligence
• Towards more Practical Threat Models in Artificial Intelligence Security
• Securing AI Systems: A Guide to Known Attacks and Impacts
• Exploiting AI for Attacks: On the Interplay between Adversarial AI and Offensive AI
• AI Agents Under Threat: A Survey of Key Security Challenges and Future Pathways