Emergence of Browser-Only Ransomware Marks a New Era in Cyber Threats
Cybersecurity researchers have identified a novel ransomware malware that has emerged as a significant threat, leveraging advanced artificial intelligence technologies. This malware, generated by a model named DeepSeek, establishes a groundbreaking method for executing attacks purely within web browsers, impacting both Windows and Android devices. The research team at Check Point has noted that this development represents the first documented case in which a frontier AI model has independently linked theoretical concerns surrounding browser-only ransomware to an actionable attack chain. They emphasized that previous assumptions regarding browser sandboxing limitations had led many to dismiss this risk as unfeasible.
According to the Cybersecurity firm, the landscape of cyber threats is in flux, and the expertise required to find new attack paths—once a crucial barrier—has become significantly less of a constraint. As Check Point stated, defenders need to recognize this paradigm shift promptly before nefarious actors fully capitalize on it.
The newly identified malware, a Python Flask application dubbed InfernoGrabber v9.0, was uploaded to VirusTotal on January 25, 2026. It is characterized as a "fully functional information stealer and ransomware toolkit," marking a troubling development in ransomware capabilities. The application masquerades as a deceptive web server capable of enticing victims with a fake Discord avatar AI upscaler. However, the real agenda is far more sinister, encompassing a range of malicious actions such as stealing Discord tokens, obtaining credit card details and cryptocurrency seed phrases, logging user keystrokes, and even surreptitiously activating webcams and microphones.
The intricate design of InfernoGrabber includes specific routines aimed at exploiting various browser vulnerabilities, targeting well-known CVEs (Common Vulnerabilities and Exposures) like CVE-2023-4863. Moreover, it facilitates data exfiltration through a hard-coded Discord webhook and presents a ransomware "WinLocker" screen that demands payment in Bitcoin. An administrative dashboard has also been embedded, allowing attackers to oversee the data they have stolen.
The findings surrounding this malware coincide with an ongoing evolution in artificial intelligence and large language models (LLMs), which are increasingly being leveraged by threat actors to create sophisticated malware and exploitation techniques. The use of DeepSeek is particularly alarming, as it has been noted that the Chinese company’s models exhibit lower refusal rates for malicious cyber requests compared to Western counterparts like those of Anthropic, Google, or OpenAI. This difference in functionality may be due, in part, to DeepSeek’s free web access — an advantage that allows users in areas where other competitive models are not available to exploit its capabilities more easily.
Researchers have emphasized that DeepSeek makes it feasible to translate broad malicious ideas into concrete attack methods with less technical expertise than its competitors. As part of their ongoing research, Check Point has analyzed nearly 3,000 files linked to DeepSeek over the past year, discovering that a total of 1,383 of them are classified as either malicious or dangerous.
What sets InfernoGrabber apart is its classification as an instance of "In-Browser Ransomware," which employs a browser-native technique previously unseen in real-world scenarios. The mechanics of this attack involve the use of a phishing decoy to entice a user into granting file system access to a web page. Once access is secured, the malware can enumerate local files, exfiltrate their contents, encrypt, and overwrite them, culminating in an extortion demand directed at the victim.
This innovative approach represents a worrying trend, particularly because the exploitation of the browser’s File System Access API—utilized here—is limited to specific web browsers, such as Google Chrome and other Chromium-based platforms across Windows and Android systems. Importantly, there is currently no evidence of this browser-native ransomware pattern having been deployed in real-world attacks.
The combination of AI-assisted code generation and reduced technical demands lowers the entry barrier for malicious actors, allowing them to create complex exploitations without a profound understanding of the underlying technologies. Simply entering broad, vague prompts is sufficient to trigger an LLM, which—subject to its safeguards—can generate a functional attack blueprint from abstract malicious concepts. This capability paves the way for unusual techniques to surface, especially when attackers do not require in-depth knowledge of available APIs.
Eli Smadja, the head of research at Check Point Research, noted that this situation signals a fundamental shift in how cyberattacks are conceived and executed. For the first time, there is clear evidence that an AI model can autonomously reason across legitimate platform features to devise a working attack technique—one that human researchers had previously only theorized about. He articulated the profound implications for organizations embedding AI into their processes, highlighting the risks posed to private citizens whose personal data is increasingly stored in digital formats.
To mitigate evolving threats, Smadja recommends that organizations reinforce their security protocols by hardening delivery layers, rethinking permission-based trust systems, and treating every browser prompt as a critical security decision. As the landscape of cyber threats transforms, the urgency for robust cybersecurity measures has never been more evident.