llm-security

Explore articles and tutorials about llm-security on Victor Johnson's Blog.

2 articles

Can AI Agents Hide Backdoors? Anthropic Research Reveals a New Data Poisoning Threat

29 October 2026

New Anthropic-linked research shows how poisoning just 0.1% of AI training data can create hidden backdoors in coding agents that survive clean training.

AI Distillation Attacks Explained - Why Anthropic Is Taking Model Theft Seriously

14 October 2026

Anthropic's September 2026 report reveals large-scale AI distillation attacks. Learn how model extraction works and why it raises security concerns.