As artificial intelligence becomes deeply integrated into the operational fabric of modern enterprise, a critical new vulnerability has emerged: the susceptibility of AI systems to sophisticated ransomware attacks. In response to this evolving threat landscape, cybersecurity leaders now have access to a specialized, free, and open-source resource designed to identify and mitigate the weaknesses that could compromise their AI infrastructure.
The tool, known as the Ransomware Assessment and Incident Defense Framework for AI Systems (RAID-AI), is now fully available for public use. It represents the culmination of two years of intensive development by Behnaz Karimi and Yuvaraj Govindarajulu, the co-leads of the Open Worldwide Application Security Project’s (OWASP) AI Bill of Materials (AIBOM) initiative. The project was initially slated for an official unveiling at the ISC2 Security Congress held in Denver, Colorado, in late October; however, unforeseen visa challenges prevented Karimi from attending the event in person. Despite this hurdle, the framework’s release marks a significant milestone in providing organizations with a structured methodology to harden their AI environments against the growing tide of digital extortion.
A Manual Ransomware Readiness Assessment for AI Systems
In an interview with Infosecurity, Behnaz Karimi clarified that RAID-AI is designed to address the unique intersection of AI and ransomware. The framework is platform-agnostic, meaning it is designed to defend against attacks on AI systems regardless of whether the adversary is utilizing AI-enhanced tools to conduct their operations.
The philosophy behind RAID-AI is distinctly hands-on. Rather than requiring users to plug into a third-party scanning tool or connect their sensitive systems to an external cloud platform—which could introduce its own security risks—the framework functions as an internal assessment tool. It compels security teams to rigorously evaluate their own controls, fostering a deeper understanding of their security posture rather than relying on automated black-box diagnostics.
The urgency of this initiative is underscored by a shifting threat environment. As Karimi noted, the industry is witnessing a clear upward trend in ransomware attacks specifically targeting AI infrastructure. The landscape is increasingly fraught with sophisticated campaigns, such as the data poisoning techniques highlighted in an April 2025 report by Zoran Mitrovic, head of engineering at Kintra Health. Furthermore, high-profile incidents, such as the alleged June 2026 cyber-attack on pharmaceutical giant Novo Nordisk—where hackers claimed to have successfully exfiltrated proprietary AI models—serve as stark reminders that intellectual property and machine learning assets are now primary targets for ransomware actors.
The RAID-AI assessment is comprehensive, covering standard cybersecurity hygiene while simultaneously addressing the specialized risks inherent to AI models, training data, and the supporting infrastructure that maintains these systems. By providing a structured web console, the framework is tailored for CISOs, cybersecurity officers, and security engineers who need to translate high-level security goals into actionable, measurable outcomes.
Users are tasked with assessing their internal controls manually, assigning maturity levels on a scale from zero to five. These controls are logically segmented into general cybersecurity categories—such as organizational recovery, staff training, and sensitive data protection—and AI-specific categories, which include data integrity, tooling, and runtime security. The framework categorizes these controls across levels ranging from basic to advanced and high, allowing organizations to map their progress against recognized industry benchmarks.
Crucially, RAID-AI aligns itself with existing, well-regarded security standards. The framework maps its controls to the US National Institute of Standards and Technology’s (NIST) Cybersecurity Framework 2.0 (CSF 2.0) and the US Cybersecurity and Infrastructure Security Agency’s (CISA) Cybersecurity Performance Goals 2.0 (CPG 2.0). By bridging the gap between general IT security and AI-specific risks, it allows organizations to integrate AI security into their broader governance programs. Furthermore, the tool enables users to map potential adversarial attacks against established frameworks including MITRE ATT&CK, MITRE ATLAS, and the NIST Adversarial framework. To support this implementation, Karimi and Govindarajulu have provided extensive documentation, including detailed guidelines on how to utilize the tool and the mathematical logic behind its maturity scoring system.
Future Plans: Extending Ransomware Assessments to Agentic AI Systems
A cornerstone of the RAID-AI design is its commitment to data privacy and organizational autonomy. Because the assessment does not require users to connect their internal systems to the internet or upload sensitive company documentation, it avoids the risks associated with data leakage.
The framework is explicitly not an automated audit. Instead, the results are a reflection of the organization’s own honest evaluation of its security maturity. The tool does not perform independent verification, nor does it claim to provide an exhaustive penetration test. Rather, it serves as a diagnostic roadmap. Karimi emphasizes that the most effective use of the tool involves bringing security managers, software engineers, and AI-security specialists together in collaborative workshops to conduct the assessment. This multi-disciplinary approach ensures that the resulting security strategy is rooted in both the technical realities of the AI models and the strategic goals of the enterprise.
Once the assessment is complete, the RAID-AI console generates interactive dashboards that visualize the organization’s maturity levels and highlight specific areas of deficiency. These dashboards are designed to serve two primary audiences: senior leadership, who require a high-level overview of risk, and technical teams, who need the granular data necessary to implement specific remediation tasks. Users can download comprehensive PDF reports, which serve as essential documentation for leadership teams to prioritize security investments and address identified vulnerabilities.
Looking ahead, the development team has ambitious plans for the future of the framework. Karimi noted that she intends to build out a repository of practical implementation resources. These will include video tutorials and technical blueprints designed to assist security engineers in the day-to-day application of individual security controls.
Perhaps most significantly, the team is looking toward the next frontier of AI security: agentic AI systems. As AI agents gain the ability to autonomously perform tasks and interact with enterprise environments, the potential for ransomware to exploit these autonomous processes grows exponentially. By developing a specialized version of the RAID-AI tool for agentic systems, Karimi and Govindarajulu aim to ensure that as AI becomes more capable and autonomous, the security community remains at least one step ahead of the adversaries seeking to exploit these technologies for financial gain. For now, the current RAID-AI framework stands as a vital resource for any organization currently navigating the complexities of securing their AI investments against the relentless pressure of the modern ransomware threat.
