Combatting scalable AI attacks
summary
Argues that generative models turn manual reconnaissance and message crafting into automated pipelines that scale to thousands, enabling LLMs generate tailored phishing and automation chains discover vulnerabilities. Recommends combining behavioral detection, provenance signals, and rate limiting; demonstrates a pipeline that feeds model outputs through content classifiers and anomaly detectors and blocks simulated attacks in testing. Emphasizes proactive model hardening, monitoring model outputs for misuse, and integrating detection closer to data sources rather than relying only on perimeter controls.
AI-written summary · about
related talks
49mThe Intersection of AI and Cybersecurity
49mGenAI and Cybersecurity: Safeguarding the Digital Age
46mGenerative AI: Cyber resiliency and risk
18mCommit Secure AI: Embedding Security in Code, Pipelines, and Production
43mNavigating the AI Frontier: Investing in AI in the Evolving Cyber...
31mFighting Digital Threats in the Age of AI
42mGenerative Email Attacks and How to Defend Against Them
28mRound Table | Generative AI: Threat or Ally?