Browse by Topic
146 posts across 8 subject areas — navigate AI security like a curriculum.
Defense Patterns
Practical mitigations: threat modeling, red-teaming methodologies, monitoring pipelines, incident response playbooks, and least-privilege architectures for AI systems.
94 postsAgent Security
Risks unique to autonomous LLM agents: tool misuse, multi-agent trust, goal hijacking, and resource exhaustion. This is the frontier of AI-specific attack surface.
63 postsPrompt Injection
Direct and indirect prompt injection, jailbreaks, system-prompt leakage, and instruction-override attacks — the most exploited class of LLM vulnerability.
50 postsLLM & Model Security
Attacks targeting the model itself: fine-tuning vulnerabilities, RAG poisoning, model extraction, weight theft, and inference-time manipulation of large language models.
48 postsAI Safety & Alignment
Reward hacking, specification gaming, RLHF failure modes, and the gap between intended and learned behavior — where safety research meets security practice.
32 postsSupply Chain Attacks
Backdoors in training data, model weights, and third-party components. Sleeper agents, poisoned checkpoints, and dependency confusion in AI pipelines.
20 postsAdversarial ML
Evasion attacks, membership inference, data poisoning, and adversarial examples — classical adversarial machine learning applied to modern foundation models.
20 postsPrivacy & Data Security
Differential privacy guarantees, gradient leakage in federated learning, data exfiltration via model outputs, and privacy-preserving training techniques.
18 posts