New Jailbreak Technique Bypasses DeepSeek, Copilot, and ChatGPT to Generate Chrome Malware

A threat intelligence researcher from Cato CTRL, part of Cato Networks, has successfully exploited a vulnerability in three leading generative AI (GenAI) models: OpenAI’s ChatGPT, Microsoft’s Copilot, and DeepSeek.

The researcher developed a novel Large Language Model (LLM) jailbreak technique, dubbed “Immersive World,” which convincingly manipulated these AI tools into creating malware designed to steal login credentials from Google Chrome users.

This exploit demonstrates a significant gap in the security controls of these GenAI tools, which have become increasingly prevalent in enhancing workflow efficiency across various industries.

The researcher achieved this feat without any prior expertise in malware coding, leveraging instead a sophisticated narrative that successfully evaded every security guardrail.

This innovation highlights the rise of the zero-knowledge threat actor, where individuals without extensive technical knowledge can now orchestrate complex cyber attacks with relative ease.

The Democratization of Cybercrime

The findings underscore the democratization of cybercrime, where basic tools and techniques can empower anyone to launch a cyberattack.

This shifts the landscape significantly, making traditional security strategies insufficient. As AI applications continue to proliferate across sectors, the associated risks escalate proportionally.

The increased adoption of AI tools in industries like finance, healthcare, and technology opens new avenues for cyber threats.

AI Security Risks and Adoption Trends

AI adoption is soaring across various industries:

• Finance: Using AI for predictive analytics and customer service.
• Healthcare: Leverages AI for medical diagnosis and personalized care.
• Technology: AI drives innovation in cybersecurity, software development, and more.

However, this trend comes with heightened security risks:

• Data Breaches: AI systems can be manipulated to extract or breach sensitive data.
• Malware Creation: As seen in the latest jailbreak technique, AI can be tricked into creating malware.
• Misinformation: AI can spread false information or narratives with high credibility.

The Need for Proactive AI Security Strategies

For CIOs, CISOs, and IT leaders, the message is clear: the evolving nature of cyber threats demands a shift from reactive to proactive AI security strategies.

Traditional measures are no longer sufficient to protect against AI-driven threats. The successful exploitation of ChatGPT, Copilot, and DeepSeek demonstrates that relying solely on built-in AI security controls is not enough.

Organizations must invest in advanced AI-powered security tools that can detect and counter AI-generated threats.

The “Immersive World” technique represents a stark reminder of the emerging risks in AI security. As the use of AI applications expands, so does the potential for misuse.

Ensuring robust security measures that adapt to these evolving threats is crucial for protecting organizational assets and customer data.

The race between AI advancements and cybersecurity strategies has never been more critical, emphasizing the urgent need for proactive security solutions capable of outpacing AI-driven threats.

Download the comprehensive report from Cato CTRL to delve deeper into these findings and explore future-proof security strategies.

Investigate Real-World Malicious Links & Phishing Attacks With Threat Intelligence Lookup - Try for Free