Flashpoint counted 6,256 ransomware attacks in the first half of 2026, a 45% jump over the previous period, as Ransomware-as-a-Service operations expanded and operators leaned harder on AI tooling. At the same time, a new open framework landed to help defenders assess whether their own AI systems can survive the same treatment.
The Numbers And The Sectors
The technical reality is stark. Manufacturing took 18% of observed hits, business services and consulting 13%, healthcare 9%. Groups such as Qilin and Akira keep demonstrating the scale at which RaaS affiliates can operate when the barrier to entry keeps falling. Flashpoint’s guide, drawn from its primary-source intelligence, maps the full lifecycle from reconnaissance through encryption and negotiation, and where intelligence on compromised credentials, vulnerability exploits and underground chatter can interrupt the chain.
AI On Both Sides Of The Wire
AI is reshaping the economics of the attack. Commoditised tooling lets operators automate phishing, credential stuffing and vulnerability scanning at speed. But the same technology gives defenders visibility across customer fleets and underground forums that no single government can match. The detail that matters is the gap between them: frontier labs see malicious patterns across services; governments see attacks on critical infrastructure. Neither has the complete picture, and ransomware is where that blind spot hurts most.
A Framework Built For The Model, Not The Perimeter
Enter RAID-AI, the Ransomware Assessment and Incident Defense Framework for AI Systems. Developed over two years by Behnaz Karimi, co-lead of the OWASP AI Bill of Materials initiative, with Yuvaraj Govindarajulu, the framework is free, browser-based and deliberately manual. Security teams score their own controls from zero to five across general categories (recovery, training, sensitive-data protection) and AI-specific ones (data, tools, runtime). No agents, no scans, no uploaded documents. The console maps results to NIST CSF 2.0, CISA CPG 2.0, and three adversarial mappings: MITRE ATT&CK, MITRE ATLAS and NIST Adversarial.
Why Now
Karimi told Infosecurity that ransomware attacks targeting AI systems are slowly becoming visible. She pointed to an April 2025 report by Zoran Mitrovic, head of engineering at Kintra Health, detailing data-poisoning techniques used by ransomware actors, and an alleged June 2026 breach at Novo Nordisk where attackers claimed theft of pharmaceutical AI models. The framework was meant to debut at the ISC2 Security Congress in Denver in late October; visa issues kept Karimi away.
The Workflow
As ever, the tool is a workshop aid, not an audit. Karimi wants security managers, engineers and AI specialists in the same room scoring controls together. Dashboards give senior leadership a maturity view; the underlying controls guide technical remediation. PDF reports highlight weaknesses and priority setting. The next step on the roadmap: extend the assessment to agentic AI systems, where autonomy expands the attack surface further.
RAID-AI is live at raid-ai.org.
Sources
- Ransomware in the Age of AI: How Defenders Can Break the Attack Chain
- List of cyberattacks
- Ransomware in the Age of AI: How Defenders Can Break the Attack Chain
- Asian Cybersecurity in the Age of AI and Quantum Computing
- DXC’s Xchanging Subsidiary Falls Prey to Ransomware Attack
- New Framework Puts Ransomware Resilience for AI Systems in Focus
Photo: Julio Lopez / Unsplash (Unsplash License)
