WormGPT Infestation

Just like everyday consumers use AI tools such as ChatGPT to automate tasks and summarize information, cybercriminals are also experimenting with their own AI-driven tools.

One example is WormGPT, a tool advertised since 2023 on underground cybercrime forums as an uncensored AI assistant designed for malicious use.

About WormGPT

WormGPT is the name used for an uncensored large language model (LLM) advertised on underground forums in 2023. It was promoted as a tool designed to generate phishing emails, business email compromise content, and basic malicious scripts without the safety restrictions found in mainstream AI platforms.

Unlike consumer AI services such as OpenAI’s ChatGPT, Google’s Gemini, or Anthropic’s Claude, WormGPT was marketed specifically to remove ethical guardrails and content filtering.

Cyber Researchers analyzing early versions reported that WormGPT was likely built on top of open source models such as GPT-J rather than a proprietary system comparable to GPT 3. There is no verified evidence that it was architecturally equivalent to GPT 3. Instead, it appears to have been a fine tuned open source model optimized for generating convincing phishing and fraud content.

Timeline and Emergence

WormGPT surfaced in 2023 on cybercrime and breach forums. It was marketed toward:

• Novice threat actors

• Business email compromise operators

• Fraud groups

• Low skill malware developers

It was positioned as a subscription service. Yes, there is a market for this.

Cybercrime has “Security” as a Service (SaaS) economics now. That alone should make every executive sit up straighter.

In other words… it’s a cybercriminal’s Multi-level Marketing.

Since its initial exposure, multiple variants and rebrands have appeared. Some disappear quickly, others resurface under new names. The underground ecosystem moves fast and rebrands often. Just like the market we know, tech moves fast and is catered to its audience.

Key Risk Factors

1. Lower barrier to entry.

   AI generated phishing removes the need for strong writing skills. Grammar errors used to be a detection clue. That characteristic is fading and being replaced.

2. Scale and personalization.

   Threat actors can generate tailored lures at scale, increasing success rates.

3. Automation of low grade malware scripting.

   While not capable of producing highly advanced nation state grade malware independently, these tools can assist with boilerplate code and modification.

Important Clarifications

• There is no verified public evidence that WormGPT “learns” dynamically from live user input the way commercial AI platforms retrain models. Most likely it is a static fine tuned model with periodic updates.

• It is not confirmed to be technically comparable to GPT 3 level performance.

• It does not represent a fundamentally new attack class. It accelerates existing social engineering and fraud operations.

In short, it is a subscription to let anyone become a low class cybercriminal. This is not a new malware class, but allows anyone to become part of the existing cybercriminal problem.

What This Means for Your Business

1. Evaluate Detection Capabilities.

Work with your security vendor to determine whether:

• Email security tools analyze linguistic patterns beyond grammar.

• Behavioral detection is in place for anomalous sending patterns.

• BEC detection includes contextual and transactional analysis.

As of today, there are checks for detecting AI generated content. However, this is not a reliable primary defense strategy. AI detection tools produce false positives and false negatives. These are roadblocks even university plagiarism tools are having. For a cyber program, it is recommended to focus on behavior and intent, not whether AI was used.

2. Disable or Restrict Macros and Script Execution

• Disable macros by default in Microsoft Office.

• Enforce least privilege.

• Use application control where possible.

AI makes phishing cleaner. Macros still execute the payload. Kill the execution path.

3. Reinforce Security Awareness

Remind employees:

• Help desk will never ask for passwords.

• Financial changes must go through verified purchase portals.

• Vendor payment updates require out of band confirmation.

Shift training away from “spot the typo” and toward “verify the transaction.”

Executive Takeaway

WormGPT and similar models are not creating a whole new era of terrible threats, yet. Understanding and becoming educated with this cybercriminal tools can provide awareness towards automated campaigns. Behind the scenes, it is commercialization of uncensored open source LLMs for fraud and phishing.

The real risk is operational efficiency for criminals.