Full Breakdown
AI-Driven Transformation of Internet Security: Defense, Threats, and Policy
6/4/2026, 11:57:54 AM
AI-Integrated Network Defense Initiative
Wayne State University’s Dr. Rhongho Jang, an assistant professor of computer science, received a five-year, $599,998 NSF CAREER award to embed artificial-intelligence (AI) directly into core network routers. The project aims to shift threat analysis from distant cloud data centers to the traffic flow itself, creating what Jang describes as a “smart immune system for the internet.” By processing traffic locally, the system can diagnose attacks instantly rather than sending samples to a remote “hospital” for analysis, potentially reducing outage time and extending real-time protection to ordinary users.
Surge in AI-Powered Offensive Tools
Researchers at the University of Toronto demonstrated a prototype worm that uses open-source AI models to locate and exploit any known software flaw across Linux, Windows, and IoT devices without human intervention. The worm harvests passwords, siphons processing power for further reasoning, and can retarget a machine after a patch is applied. In parallel, cyber-crime analyst Cynthia Kaiser reported a 3 810 % rise in AI-enabled illicit tools on underground markets—from 38 listings in December to 1 486 in February. The FBI’s 2025 Internet Crime Report recorded 22 364 AI-related complaints, costing victims more than $893 million.
Scale of AI-Driven Cybercrime
The FBI logged over 3 000 internet-crime complaints per day in 2025, surpassing one million annual reports and $20 billion in losses. AI-driven scams accounted for $893 million of those losses, while investment fraud—often amplified by AI-generated deepfakes and synthetic voices—totaled $8.6 billion. Phishing, spoofing, and AI-powered chatbots that impersonate corporate executives remain the most frequent vectors, highlighting the breadth of AI-enabled abuse.
Official Responses and Industry Perspectives
- Defense research: Dr. Jang emphasizes that recent AI advances make it feasible to move defensive capabilities “beyond academic exploration and toward real-world deployment.”
- Threat awareness: University of Toronto lead author Nicolas Papernot warned that “once a worm is launched, the cost would drop to nearly zero,” urging rapid collaboration among researchers, industry, and policymakers.
- Law-enforcement stance: The FBI’s report stresses that AI-driven cybercrimes are “exploding” and calls for heightened vigilance across sectors.
- Market-level mitigation: Kaiser urged coordinated action, noting that “effective disruption of this market requires coordination between defenders, model providers, payment processors and hosting infrastructure.”
Critiques and Concerns Over Dual-Use AI
Experts caution that AI alignment mechanisms, while steering models toward factual (“logos”) responses, strip away the “ethos” and “pathos” cues humans rely on to assess intent. Kevin Esterling (UCR) observed, “Humans essentially use all three, whereas LLMs essentially only rely on logos.” This reduction in nuanced reasoning may make social-engineering attacks harder to detect and could erode trust in online discourse, compounding security challenges.
Verbatim Quotes
- “We need to catch up and find ways we can proactively use AI to create a more secure environment.” — Rhongho Jang, Assistant Professor, Wayne State University
- “In an interconnected world, no system is immune to this threat,” — Nicolas Papernot, Lead Author, University of Toronto
- “Effective disruption of this market requires coordination between defenders, model providers, payment processors and hosting infrastructure “The good news is that the same intelligence work that reveals how these markets operate also reveals where they’re vulnerable,” Kaiser concluded.” — Cynthia Kaiser, SVP, Halcyon
- “Those crimes accounted for more than $215 million in losses.” — Federal Bureau of Investigation (2025 Internet Crime Report)
- “What we found is that humans essentially use all three of those, whereas LLMs essentially only rely on logos,” — Kevin Esterling, Professor of Public Policy, UCR
Outlook: Research, Regulation, and Resilience
Jang’s NSF-funded effort plans to prototype AI-enhanced routers by 2028, targeting widespread deployment in enterprise and consumer networks. Simultaneously, the FBI recommends public-private information-sharing frameworks to track AI-enabled threats. Kaiser calls for policy reforms that align model providers with law-enforcement priorities, while Papernot urges the research community to publish defensive countermeasures alongside offensive demonstrations. Together, these actions aim to balance AI’s promise for faster, smarter internet infrastructure with safeguards against its misuse.
