Explore

Collections & Series

Browse research by topic, paper trail, and recurring adversarial ML themes.

5 posts / llm-jailbreaks-defenses

LLM Jailbreaks & Defenses

Automated jailbreaks, prompt attacks, and the defenses built to catch or contain them.

3 posts / adversarial-optimization

Adversarial Optimization

How search, evolution, gradients, and black-box optimization discover failures in ML systems.

2 posts / supply-chain-security

Supply Chain Security

Security risks in the AI software stack, from model dependencies and agent tools to package ecosystems, plugins, and deployment pipelines.

2 posts / autonomous-agents

Autonomous Agents

Security risks that emerge when AI systems can plan, call tools, write code, use memory, coordinate tasks, or act with partial autonomy.

1 posts / computer-vision-attacks

Computer Vision Attacks

Attacks against image classifiers, perception systems, and vision-model assumptions.

1 posts / reinforcement-learning-security

Reinforcement Learning Security

How reward functions, agents, environments, and evaluators can be gamed, exploited, or misaligned under optimization pressure.

1 posts / ai-security-fundamentals

AI Security Fundamentals

A ground-up series on why AI systems are vulnerable by design, not by accident. Starting from the decision boundary every model relies on, this series builds the theory needed to understand every major AI attack category, and what can actually be done about each one.