Got 99 Problems but Prompt Injection Ain't Watermelon
Navigating the complexities of responsible vulnerability disclosure, data bias, and model manipulation in AI systems.
Summary
Vulnerability Disclosure in Artificial Intelligence
Disclosing security flaws in artificial intelligence introduces distinct challenges compared to traditional software. Model non-determinism, proprietary weights, and data privacy complicate remediation and reproduction.
Data Bias and Model Manipulation
Security risks in AI extend beyond prompt injection to include data poisoning, training bias exploitation, and model tampering, which can systematically distort system outputs without triggering traditional alarms.
Building Standardized AI Disclosure Frameworks
Establishing transparent, responsible disclosure standards across AI vendors, security researchers, and regulatory bodies is essential for maintaining public trust and system integrity.
Related talks
App Sec Village·27 min
Javan Rasokat · Andra Lezza
Practical insights and defensive strategies from real-world development and hardening of production LLM applications.
Briefings·26 min
Jiashuo Liang · Guancheng Li
Introduces a novel gradient-based prompt-injection technique that can generate universal triggers to manipulate open-source Large Language Model (LLM) outputs.
CyberRisk TV·16 min
Ramin Farassat
Ramin Farassat investigates How Menlo Security Is Securing AI Agents from Prompt Injection.