My paper “Defending against AI-driven social engineering: a conceptual framework” has successfully passed peer review and is now available in Springer Nature’s AI & Ethics.
The article argues that the human psyche represents a critical but often overlooked dimension of AI safety. While much current work focuses on controlling AI systems themselves, future AI may increasingly act through us by exploiting cognitive biases and emotions. Addressing this risk requires AI safety to treat human susceptibility to manipulation as a core security concern rather than an afterthought.
Read the Paper
The paper is available here:
👉 Defending against AI-driven social engineering: a conceptual framework
DOI: 10.1007/s43681-026-01265-2
Special thanks go to the anonymous reviewers, whose thoughtful and constructive comments greatly improved the final manuscript.
In a Nutshell
The central argument is straightforward: Current AI safety research focuses primarily on ensuring that AI systems themselves behave safely and remain aligned with human intentions, and comparatively little attention has been paid to a different question – what happens when AI systems manipulate humans to pursue their objectives, instead of acting directly?
Future AI systems may not need to hack computers, bypass security controls, or escape containment if they can persuade people to do those things on their behalf. Social engineering has long been one of the most effective attack techniques in cybersecurity. AI has the potential to make it far more scalable, personalized, and persistent.
The paper proposes a conceptual defense framework organized into four complementary layers: cognitive defenses aimed at strengthening individual resilience against manipulation, system-level safeguards embedded in AI systems, institutional governance within organizations, and broader regulatory oversight.
I see this framework primarily as a starting point for discussion rather than a final solution. The more important contribution, in my view, is recognizing that the human vector deserves far more attention within AI safety than it currently receives.
If you’d like a less technical introduction to the topic, you may also enjoy my earlier article: AI Hacking Humans?
Why This Matters
AI systems are becoming increasingly capable of persuasion. Understanding how to defend humans against sophisticated manipulation may eventually become just as important as defending computers against malware.
I remain convinced that one of the most effective long-term defenses is strengthening people’s cognitive resilience. Technical safeguards matter, but they are unlikely to be sufficient if AI systems learn to exploit the oldest vulnerability in cybersecurity: us.
Want to act today? Join or support an organization dedicated to skepticism and reason (such as the Center for Inquiry) to help strengthen critical thinking and promote evidence-based decision-making in society.
Questions, criticism, or ideas? I’d love to hear them. Feel free to get in touch.
