On the Feasibility of Using MultiModal LLMs to Execute AR Social Engineering Attacks

This is an academic research paper (0 stars, 356 days old, 0 velocity) that presents the SEAR framework—a proof-of-concept combining AR technology with multimodal LLMs to execute social engineering attacks. Key threat signals: (1) **Defensibility is minimal**: It's a reference implementation accompanying a published paper with no production deployment, no user base, and no proprietary moat. The novelty lies in combining existing technologies (AR + multimodal LLMs) rather than introducing new primitives. (2) **Platform domination risk is HIGH**: OpenAI (GPT-4V), Google (Gemini, ARCore), and Meta (with AR/VR investments) all control the core components—the multimodal models and AR frameworks. They can trivially integrate guardrails, detection, or security features into their platforms within months. The attack surface itself becomes a defensive opportunity for these platforms. (3) **Market consolidation risk is LOW**: There is no incumbent market for 'AR social engineering attack frameworks'—this is a security research contribution, not a commercial product. No acquisition target exists because the value is in the research insight (which is now published), not the code. (4) **Displacement horizon is 6 months**: Platform vendors (OpenAI, Google, Anthropic) are actively hardening multimodal models against misuse. Detection and mitigation of AR-augmented social engineering will be incorporated into their safety roadmaps immediately. The research finding is already exposed; implementation becomes a liability rather than an asset. (5) **Integration surface is theoretical/reference**: The code is academic scaffolding to demonstrate a concept, not a composable component. It relies on APIs and frameworks controlled by dominant platforms. (6) **Novelty is novel_combination**: The insight is valuable for security research, but it's not a new technique—it's a demonstration that existing capabilities (multimodal understanding + AR + social engineering) combine to create a new threat. The defensibility of such research is inherently limited because the contribution is the *idea*, not an unreplicable implementation. The 11 forks suggest academic interest, not commercial adoption. This project has no commercial defensibility and minimal technical moat. Its value as a threat intelligence artifact is high for defensive players (platform vendors), which is why displacement is imminent.