Microsoft Names Developers Behind AI Jailbreaking Tools in Legal Crackdown on Storm-2139

In a landmark move against AI exploitation, Microsoft has unmasked key developers behind Storm-2139, a global cybercrime network accused of manipulating generative AI models for illicit activities. The company’s legal action targets four named defendants who allegedly bypassed safety measures in Microsoft’s Azure OpenAI Service and other AI platforms, reselling unauthorized access to bad actors worldwide.

A Global Network of AI Exploiters

Microsoft’s Digital Crimes Unit (DCU) has been investigating Storm-2139 since December 2024, originally filing a lawsuit against ten unidentified “John Does” in the Eastern District of Virginia. The updated complaint now identifies four individuals:

• Arian Yadegarnia (aka “Fiz”) of Iran

• Alan Krysiak (aka “Drago”) of the United Kingdom

• Ricky Yuen (aka “cg-dot”) of Hong Kong

• Phát Phùng Tấn (aka “Asakuri”) of Vietnam
  

According to Microsoft, these individuals played pivotal roles in developing and distributing tools that enabled the circumvention of AI safety guardrails. By exploiting stolen credentials from publicly available sources, Storm-2139 provided unauthorized access to generative AI systems, allowing users to generate non-consensual intimate images of celebrities and other illicit content.

“As organizations adopt AI tools to drive growth, they also expand their attack surface with applications holding sensitive data,” said Rom Carmel, Co-Founder and CEO at Apono. “To securely leverage AI and the cloud, access to sensitive systems should be restricted on a need-to-use basis, minimizing opportunities for malicious actors.”

The Storm-2139 Playbook: From Creators to Users

Microsoft’s investigation reveals a structured operation with three primary roles:

• Creators: Developed AI jailbreaking tools that override built-in safety measures.

• Providers: Distributed and monetized these tools, offering access tiers and subscriptions.

• Users: Purchased illicit access to manipulate AI systems for generating explicit content.
  

The network’s operations extended across multiple countries, with actors in the United States, Austria, Russia, India, China, and beyond. Microsoft has referred its findings to both U.S. and international law enforcement agencies for potential criminal proceedings.

AI Hijacking: A Growing Cybersecurity Threat

Security experts warn that AI-enabled cybercrime is evolving rapidly. Patrick Tiquet, Vice President, Security & Architecture at Keeper Security, highlighted the rise of “LLMJacking,” a term referring to the hijacking of large language models via stolen credentials.

“Storm-2139’s exploitation of exposed API keys to hijack GenAI services underscores the need for robust credential hygiene and continuous monitoring,” Tiquet explained. “Attackers not only resold unauthorized access but actively manipulated AI models to generate harmful content, bypassing built-in safety mechanisms.”

To mitigate risks, experts recommend enforcing least-privilege access, implementing strong authentication, and securely storing API keys in digital vaults. Regular credential rotation, anomaly detection, and automated threat monitoring are critical measures to defend against similar attacks.

Cybercriminals React to Microsoft’s Legal Offensive

Microsoft’s aggressive approach has already disrupted Storm-2139’s operations. The company successfully seized a key website used by the group, prompting internal strife among cybercriminals. In monitored chat channels, members speculated on the identities of the “John Does,” with some publicly revealing personal information about fellow hackers in an attempt to deflect blame.

Meanwhile, Microsoft’s legal team became a direct target. Threat actors “doxed” company attorneys by posting personal details, photos, and contact information online—a harassment tactic commonly used by cybercriminals to intimidate individuals.

As cybercriminals scramble, Microsoft remains steadfast in its mission to combat AI abuse. Elad Luz, Head of Research at Oasis Security, emphasized the importance of securing AI service accounts from similar threats.

“In an era where AI safety is a high priority, Microsoft is taking action,” Luz said. “They have traced down and are prosecuting threat actors who are abusing stolen LLM access. Organizations must proactively secure service accounts, service principals, API keys, and other non-human identities that could serve as entry points for these types of attacks.”

The Battle Against AI Exploitation Continues

Despite these disruptions, experts caution that AI exploitation will remain a persistent threat. J Stephen Kowski, Field CTO at SlashNext, warns that LLMjacking can create a domino effect, with compromised credentials fueling widespread abuse across multiple threat groups.

“The primary danger of LLMjacking is that it creates a domino effect where initial credential theft leads to widespread abuse by multiple bad actors,” Kowski explained. “Beyond the significant financial impact from unauthorized AI usage charges, these attacks enable the creation of harmful content including sexually explicit material, bypassing the safety controls built into these systems.”

To counteract these threats, companies must implement multi-factor authentication for AI services, establish strict role-based permissions, and monitor AI-related activity for anomalous behavior. Proactive security measures such as logging API usage, detecting billing anomalies, and conducting regular audits are essential in preventing unauthorized access.

Microsoft’s legal campaign against Storm-2139 marks a turning point in the fight against AI exploitation. By exposing and dismantling malicious networks, the tech giant aims to set a precedent for the industry—one that prioritizes the ethical and secure use of AI technologies. But as Microsoft itself acknowledges, the battle against AI misuse is far from over.

“As we’ve said before, no disruption is complete in one day,” Microsoft stated. “Going after malicious actors requires persistence and ongoing vigilance.”