This research provides the first comprehensive security evaluation of Small Language Models (SLMs) against jailbreak attacks, revealing concerning vulnerabilities as these models become more widely deployed on edge devices.

• SLMs demonstrate comparable or higher vulnerability to jailbreak attacks than their larger counterparts
• Existing defense mechanisms show limited effectiveness against sophisticated attacks on SLMs
• The study reveals concerning security-privacy tradeoffs when deploying SLMs in security-critical applications
• Researchers recommend "security-by-design" approaches for future SLM development

As SLMs increasingly power edge devices like smartphones and vehicles, understanding these vulnerabilities becomes critical for protecting users and systems from malicious exploitation.

Can Small Language Models Reliably Resist Jailbreak Attacks? A Comprehensive Evaluation