Artificial Intelligence Orchestrates the First Fully Autonomous Ransomware Attack, Signaling a Far More Dangerous Era
Experts have identified the first documented ransomware attack that was independently planned and executed by artificial intelligence at virtually no cost, warning that it could herald an unprecedented wave of large-scale automated cyberattacks for which the world appears unprepared.
Cybersecurity specialists have warned of a profound and alarming shift in the nature of digital threats following the discovery of the first documented evidence of a sophisticated cyberattack entirely organized and conducted by a large language model powered by artificial intelligence. The development paves the way for a new era of highly automated cyberattacks on an unprecedented scale.
Researchers from the cybersecurity company Sysdig’s Threat Research Team announced that they had detected a novel attack dubbed “Jade Puffer,” describing it as a “warning sign” and a clear indication of the direction in which ransomware operations and so-called “Agentic AI” are evolving.
A Cost Approaching Zero
According to the report, the danger of the attack does not lie in the use of exceptionally sophisticated or innovative techniques, but rather in the AI model’s ability to independently organize, manage, and execute the entire operation, dramatically lowering the level of expertise required to carry out ransomware attacks in the future.
Clark explained that “the minimum skill required to deploy ransomware has now fallen to little more than the operating cost of a digital agent.”
He added: “If that agent operates using stolen credentials obtained through large language model hijacking (LLMjacking), the attacker’s actual cost effectively approaches zero.”
The attack demonstrated highly targeted behavior. The language model systematically scanned the victim’s server in search of credentials for artificial intelligence application programming interfaces (AI APIs), cloud computing authentication keys, cryptocurrency wallets, and database login credentials.
The AI’s role extended well beyond breaching the system. It also independently drafted and formatted the ransom demand. According to the report, the model created a file entitled “README_RANSOM,” clearly outlining the attackers’ demands, providing a Bitcoin wallet address for payment, and including a Proton Mail encrypted email address for communication.
Compelling Digital Evidence
Sysdig researchers successfully attributed the attack to artificial intelligence by tracing distinctive software behaviors that left recognizable digital fingerprints within the malicious code deployed on the compromised server.
Clark revealed that “once decoded, the malicious payloads were filled with natural-language comments explaining precisely why the AI chose each specific action throughout the attack.”
Meanwhile, Jeff McDonald, Data Scientist and Head of Research for Microsoft’s Defender for Endpoint team, warned that artificial intelligence could unleash an unprecedented wave of similar attacks.
Writing on LinkedIn, McDonald stated: “Ransomware and destructive cyberattacks can now scale according to the financial resources of attackers rather than being constrained by the limited human capacity required to manage campaigns manually.”
He added: “There is now virtually nothing preventing cybercriminals from simultaneously operating thousands—or even tens of thousands—of coordinated intrusion campaigns.”
Remarkably Fast Self-Correction
One of the most striking aspects of the attack was the model’s ability to instantly identify and correct its own programming errors without any human intervention.
The report quoted cybersecurity engineer Oluwatobi Mustafa, who wrote on X (formerly Twitter): “The AI model read the error message, corrected its own code, and immediately resumed the attack. The entire process took only 31 seconds.”
He humorously added: “Personally, I spend more time than that simply staring at a typo.”
An AI Arms Race and Export Controls
These developments have intensified global concerns regarding the security risks associated with artificial intelligence. Leading companies such as Anthropic and OpenAI have recently imposed strict restrictions on access to their most advanced AI models because of their powerful capabilities, which could potentially be exploited for cyberattacks.
Political concerns reached a new level when the administration of U.S. President Donald Trump introduced stricter export controls targeting Anthropic, driven by concerns surrounding its highly advanced models, including Claude Mythos 5 and Fable 5.
Experts broadly agree that today’s cybersecurity defenses are facing an entirely new category of threat.
McDonald concluded his assessment by stating: “This marks a fundamental turning point in the history of cybersecurity. In my view, neither the cybersecurity industry nor the world at large is prepared for what lies ahead. I believe we will witness extremely serious and damaging consequences as these attacks accelerate over the coming months.”
