Secure by Design
The Evolution of AI Security: Why Secure by Design Matters
Eresus Security Research TeamSecurity Researcher
March 26, 2025
1 min read
The AI Security Landscape
The age of generative artificial intelligence (GenAI) has arrived, and businesses are adopting the technology at bullet train speed. Protecting AI systems requires a fundamental shift in security thinking.
Understanding the Unique AI Attack Surface
Unlike traditional cybersecurity, AI systems face unique vulnerabilities:
- Data Poisoning Attacks: Attackers secretly insert harmful inputs into training data, compromising AI systems before they're deployed.
- Prompt Injection Attacks: Carefully worded inputs that override safety measures.
- Model Deserialization Attacks: When AI models are packaged (like Pickle or PyTorch files), attackers embed malicious code within them causing system compromise upon loading.
Defense in Depth
A robust defense in depth (DiD) strategy includes data validation, model monitoring, runtime protection, and incident response layers specifically customized for AI.