"Think First, Verify Always": Training Humans to Face AI Risks

Humans are the critical first line of defense against AI-enabled attacks. The paper argues that cybersecurity has focused too much on protecting devices and not enough on protecting human cognition, positioning people as "Firewall Zero" against AI-driven manipulation and misinformation.

A brief training intervention can meaningfully improve people's ability to resist AI-driven cognitive attacks. A randomized controlled trial with 151 participants found that just 3 minutes of training based on five principles (Awareness, Integrity, Judgment, Ethical Responsibility, and Transparency) produced a statistically significant 7.87% improvement in cognitive security task performance compared to untrained controls.

Simple, actionable protocols work better than passive warnings. Rather than relying on generic disclaimers, the "Think First, Verify Always" protocol gives people concrete steps to follow, suggesting that practical guidance is more effective at building human resilience against AI-enabled threats.

AI platforms should integrate human-centered security training as a standard feature. The authors recommend that generative AI systems embed the TFVA protocol directly into their platforms, making responsible AI use the default experience rather than an optional add-on.

Responsible AI requires bridging technical and human factors. The paper emphasizes that trustworthy AI systems cannot rely on technology alone; they must also invest in training people to think critically and verify information, treating human judgment as an essential security component.