You’re in the SOC. A low-severity alert is triggered by an isolated endpoint. Within 90 seconds, everything changes: files are encrypted, sensitive data is exfiltrated, a chatbot initiates the ransom negotiation, and a demand appears—calculated to match your company’s financial profile.
The attack followed no known pattern. It learned, analyzed, and adapted its behavior. No static scripts, no predefined commands. This is AI-powered ransomware. And it is no longer theoretical—it has been a documented reality since 2025.
What is AI-powered ransomware?
AI-powered ransomware is a next-generation form of malware that leverages artificial intelligence to automate, optimize, and adapt every phase of the attack chain.
Unlike traditional ransomware, which relies on fixed code and repeatable patterns, this new class of malware uses language models and adaptive logic to generate malicious scripts in real time, perform contextual analysis, and dynamically evade detection systems.
Documented case: PromptLock
In August 2025, ESET revealed PromptLock, the first proof-of-concept for autonomous ransomware powered by a local open-source LLM (gpt-oss:20b via Ollama). This malware generates Lua scripts on demand to explore the system, extract data, and encrypt files on Windows, Linux, and macOS. The ransom note is also AI-generated and adapted to the target.
PromptLock’s architecture demonstrates that LLMs can become polymorphic payload engines, defeating signature-based and rule-based detection.
Three reasons why these attacks are so dangerous
- Autonomous and silent efficiency
AI ransomware can execute attacks without human interaction, analyze the environment, identify valuable assets, and encrypt critical data—while avoiding standard detection techniques. Each execution alters its footprint, making traditional indicators of compromise unstable or nonexistent.
In PromptLock, no static payload is stored. The malicious code is synthesized in real time from embedded prompts, making each instance unique.
- Precision targeting and personalization
AI enables attackers to scrape social media, public data, and internal files to craft highly contextualized attacks. Ransomware can use NLP to detect confidential documents and prioritize what to extract before encryption.
Some campaigns even deploy deepfakes—audio or video—to reinforce social engineering at the human layer.
- Lower technical barrier for attackers
LLMs allow non-technical actors to generate and operate ransomware campaigns. Groups like GTG-5004 (tracked by Anthropic) used Claude Code to build and distribute ransomware with minimal technical skill.
Another clear example is FunkSec, a ransomware group that emerged in late 2024. FunkSec leveraged AI to build and adapt their malware, offering it through a ransomware-as-a-service model. They typically demand low ransoms (around $10,000), focusing on volume over complexity. Despite lacking advanced coding skills, their attacks were successful thanks to AI-assisted evasion and automation. This demonstrates that AI is drastically lowering the technical threshold for launching effective ransomware attacks.
This gives rise to AI-enhanced ransomware-as-a-service, where attackers rely on models to handle the complexity, enabling widespread, opportunistic campaigns.
What can be done? Three pillars for cyber defense
- Industrialize security hygiene through automation
Security posture must be continuously reduced and monitored using autonomous systems:
- Deploy self-healing systems (automated patching, rollbacks)
- Continuous attack surface management (ASM/CAASM)
- Strengthened Zero Trust architecture with dynamic segmentation
- Monitor local AI models (Ollama access, localhost LLM usage)
- Implement autonomous and deceptive defenses
Countermeasures must disrupt the learning and adaptation loops of automated threats:
- Moving Target Defense: dynamically change ports, paths, topologies
- Deceptive environments: AI-fed honeypots, data decoys, fake files
- Multi-layer behavioral detection (UEBA, XDR with fuzzy logic)
- Enhance executive visibility and oversight
Leadership needs consolidated, real-time insight into cyber risk:
- Dynamic risk dashboards powered by AI
- AI-powered attack simulations (red teaming)
- Resilience indicators (behavioral detection rate, response times, endpoint coverage)
- Train and test teams against new scenarios
- Simulated phishing and deepfake-based attacks
- Integration of AI threat tactics into IR playbooks
- Targeted threat intel (LLM misuse, prompt injection, payload generation)
AI-powered ransomware marks a shift. No longer hypothetical, it is accessible, professionalizing rapidly, and capable of exploiting human, technical, and organizational vulnerabilities at scale. PromptLock and FunkSec illustrate two converging trends: technical autonomy and commoditized malicious tooling via LLMs.
There is no silver bullet. Defending against this threat requires a strategic pivot—from reactive posture to predictive, automated, and governance-driven models.
Cyber resilience begins with awareness, from technical teams to executive boards.
Sources :
EST – ESET Research découvre PromptLock, le premier ransomware piloté par l’intelligence artificielle
MIT – 80% of ransomware attacks now use artificial intelligence
Check Point – FunkSec Ransomware – AI Powered Group
