New Framework Launched to Bolster Cybersecurity in AI Systems Against Ransomware Threats
In an increasingly digital age, cybersecurity leaders now have access to a groundbreaking framework designed specifically to address vulnerabilities in artificial intelligence (AI) systems that could make them susceptible to ransomware attacks. The newly unveiled tool, named the Ransomware Assessment and Incident Defense Framework for AI Systems (RAID-AI), aims to empower organizations to proactively identify and mitigate risks.
This innovative framework is fully accessible online, free of charge, providing a valuable resource for cybersecurity professionals eager to strengthen their defenses. Developed over the past two years by Behnaz Karimi and Yuvaraj Govindarajulu, both co-leads of the Open Worldwide Application Security Project’s (OWASP) AI Bill of Materials (AIBOM) initiative, the RAID-AI tool reflects a concerted effort to address the growing concern around ransomware targeting AI infrastructures.
The initial plan was to launch RAID-AI at the ISC2 Security Congress in Denver, Colorado, scheduled for October 26-28. However, unforeseen visa complications prevented Karimi from attending the event. Despite this setback, the framework’s introduction promises to be a timely intervention given the increasing frequency of ransomware incidents associated with AI technologies.
Focus on Ransomware Readiness
In an interview with Infosecurity, Karimi elaborated on the primary focus of RAID-AI, which addresses ransomware attacks on AI systems, irrespective of whether an attacker utilizes AI in their methods. The framework’s hands-on approach encourages security teams to conduct self-assessments of their existing controls rather than relying on automated scans or system connections. "Currently, we are witnessing a gradual uptick in ransomware targeting AI systems," Karimi noted, underscoring the urgency of the framework’s release.
Karimi pointed to alarming incidents, including a report by Zoran Mitrovic, the head of engineering at Kintra Health, published in April 2025. The report outlined how ransomware actors have increasingly leveraged data poisoning techniques to execute their schemes. Furthermore, a cyber-attack on Novo Nordisk in June 2026 involved hackers allegedly stealing AI models critical to the pharmaceutical company’s operations, further illuminating the risk landscape.
The RAID-AI framework encompasses essential cybersecurity measures while tailoring risk assessments to the unique aspects of AI models, datasets, and the supporting infrastructure. Designed for use through a web console, the tool is particularly beneficial for Chief Information Security Officers (CISOs), cybersecurity officers, and security engineers seeking to enhance their organization’s defenses.
Assessment Methodology
Users engage in a manual assessment process, choosing maturity levels for various controls on a scale of zero to five. The controls are categorized into general measures and those specific to AI, with descriptors like basic, advanced, and high. General controls include recovery mechanisms, training protocols, and protection of sensitive data. AI-specific domains revolve around data handling, tool selection, and operational runtime.
Notably, RAID-AI aligns its controls with established voluntary security frameworks, such as the US National Institute of Standards and Technology’s (NIST) Cybersecurity Framework 2.0 and the US Cybersecurity and Infrastructure Security Agency’s (CISA) Cybersecurity Performance Goals 2.0. Additionally, the tool facilitates the mapping of adversarial threats through three recognized attack frameworks: MITRE ATT&CK, MITRE ATLAS, and NIST Adversarial.
To support users, comprehensive documentation has been provided by Karimi and Govindarajulu, detailing guidelines on utilizing the framework and explaining the development process. This documentation also clarifies how maturity levels and various security scores are computed.
Future Vision: Expanding Ransomware Assessments
Importantly, organizations utilizing the RAID-AI assessment are not required to connect their internal systems or share sensitive company documents, maintaining a high level of confidentiality and security during the assessment process. It is crucial to note that RAID-AI is not an automated audit; the results depend on users’ evaluations, and the tool does not independently verify an organization’s security posture. Instead, teams are encouraged to engage in collaborative reviews of their practices and enter their assessments.
Karimi envisions a model where security managers, engineers, and AI specialists convene in workshops for more effective evaluations. The console provides visual results in the form of dashboards, showcasing maturity views and interconnections to other frameworks. Users can also generate downloadable PDF reports that pinpoint weaknesses and help prioritize remediation efforts.
Karimi expressed optimism regarding future developments for RAID-AI, including the provision of practical implementation resources such as instructional videos and blueprints tailored for engineers. Additionally, there are plans to create a similar assessment tool focused on ransomware threats targeting agentic AI systems, reinforcing a proactive stance against evolving cyber risks.
The launch of RAID-AI represents a significant step toward enhancing the resilience of AI systems against sophisticated ransomware threats, setting a new standard in the intersection of cybersecurity and artificial intelligence.