Modeling Agentic Crime: A Proposed ROSI Framework for Deepfake and AI-Enabled Financial Risk

AI-enabled fraud, including deepfake-assisted identity attacks and synthetic identity schemes, is increasing pressure on financial KYC and fraud controls. To conceptualize this shift, this paper proposes the original analytical construct of "agentic crime," defined as autonomous or semi-autonomous algorithmic illicit activity operating with minimal human oversight. The objectives of this study are to define agentic crime as a distinct threat category and develop a framework to quantify its financial risk. As a proposed model term, this agentic capability may theoretically scale and automate financial fraud, compounding the institutional attack surface. Concurrently, synthetic media, particularly deepfakes, has emerged as a major and growing tactical vector for bypassing biometric identity verification protocols. As these AI-assisted schemes evolve, traditional risk management models struggle to quantify the changing threat landscape.

Methodologically, the study extends the standard Annualized Loss Expectancy (ALE) with an "agentic amplification factor," integrating it into a Return on Security Investment (ROSI) framework to model technical, procedural, and regulatory controls. Building on standard Annualized Loss Expectancy (ALE) calculations, the methodology introduces a novel "agentic amplification factor." Rather than an established metric, this design construct proposes how automation and adversarial adaptation may theoretically alter both expected loss and control value. The framework evaluates the potential mitigating effects of technical detection, procedural workflows, and regulatory controls under these modeled assumptions. Furthermore, adjustments for underreported synthetic fraud and the temporal degradation of countermeasures against evolving evasion tactics are presented as empirical calibration choices, allowing institutions to test scenarios rather than relying on factual universals.

In conclusion, this translational research bridges theoretical criminology with applied financial modeling. By proposing a framework to convert abstract AI threats into decision-ready metrics, the presentation offers fintech institutions, risk modelers, and policymakers a conceptual tool to evaluate cybersecurity investments and regulatory design against the emerging realities of AI-enabled fraud.