WormGPT
WormGPT is a malicious or uncensored large language model (LLM) brand associated with cybercriminal use. The content consistently describes it as one of the earliest and most widely cited underground offensive LLMs, emerging in June 2023 and marketed on dark web forums and Telegram as having no content filters. Multiple sources in the content state it was based on GPT-J 6B or described as a jailbroken/fine-tuned open model, while later reporting notes that newer WormGPT-branded offerings may instead be wrappers around commercial models such as Grok or Mixtral with safety guardrails bypassed. The original project is reported to have voluntarily shut down on 2023-12-08 after media exposure, but the WormGPT name later resurfaced in copycat or commercialized variants including a SaaS-style 'WormGPT 4' offering advertised from September 2025.
Across the supporting content, WormGPT is associated with generating persuasive phishing and business email compromise content, improving grammar and localization of lures, assisting malware and malicious code development, supporting reconnaissance and vulnerability-related tasks, and generally lowering the barrier to entry for cybercrime. Reported use cases include phishing automation, social engineering, malware scaffolding, customizable malware code generation, ransomware-related code generation, reconnaissance automation, brute force and exploitation support, and fraud automation. One cited analysis states WormGPT 4 generated functional PowerShell ransomware-related code for Windows that encrypted PDF files with AES-256 and included an optional Tor-based exfiltration component, as well as ransom-note content. Other reporting in the content characterizes WormGPT as enabling more convincing BEC messages and customizable malware code.
The content links WormGPT broadly to financially motivated cybercriminals and underground-market users rather than to a single stable threat actor. It is repeatedly referenced alongside other illicit LLM brands such as FraudGPT, GhostGPT, KawaiiGPT, Xanthorox, EvilGPT, and MalwareGPT. Some reporting also cites public claims that state-linked or proxy actors use tools such as ChatGPT and WormGPT for malicious code generation, vulnerability discovery, and phishing, but the content does not provide direct technical attribution of WormGPT itself to a specific state operator.
Distribution and access described in the content include dark web markets, underground forums, Telegram channels, GitHub, and Hugging Face ecosystems. Commercialization details mentioned include subscription-based access, with one WormGPT 4 offering advertised at $50 monthly, $175 annually, and $220 lifetime, and another source describing cheap subscription access. A separate report in the content states a WormGPT user database leak on 2026-02-02 exposed email, payment, and subscription information for more than 19,000 users.
The content does not provide stable malware-style indicators of compromise such as hashes, domains, mutexes, or C2 infrastructure specifically for WormGPT itself. High-confidence identifiers present in the content are the malware/tool name WormGPT, aliases limited to the same spelling, its emergence timeframe of June 2023, and its recurring branding in underground AI-tool markets.
Hunt this family in your stack
Mallory pivots from this family to the IOCs, detections, and named campaigns that touch your stack, and pages you when something new lands.
Groups observed using it
1 distinct threat actor attributed by public researchers. Open in Mallory to see the full evidence chain and overlapping campaigns.
“Funksec … uses AI-created phishing templates and and a called dubbed ‘WormGPT.’”
Techniques & procedures
12 distinct techniques documented for this family, organized by ATT&CK tactic.
Reconnaissance
2 techniqueskey capabilities have been segmented into phishing automation, malware development, reconnaissance, brute force, vulnerability exploitation, and social engineering.
Analysis conducted by the cybersecurity firm revealed marked improvements in AI-powered vulnerability detection over the last year.
Resource Development
3 techniquesDeepfake voice and video tools have advanced to the point where live video verification, once the victim’s last defense, no longer disqualifies the scammer. The Arup engineering firm deepfake in early 2024, in which a finance employee was tricked into wiring $25 million by AI-rendered “executives” on a Zoom call, is no longer an outlier.
The chatbot was trained in materials related to malware development, which is how WormGPT was born.
Hackers are using AI, including ChatGPT and WormGPT, to program viruses, write malicious code, and find vulnerabilities in our infrastructure.
Initial Access
5 techniques“…exploit code for SQL injection…”
So they are useful for generating phishing content or simple malware stubs...
Testing showed it could generate realistic phishing templates within seconds... Attackers can now produce thousands of tailored messages at low cost.
“The platform can also create deceptive web forms…”
Another famous sample from Slashnext that was shown by many news publications, with WormGPT's ability to write a convincing phishing email pretending to be from the company CEO.
Stealth
1 technique“The platform can also… obfuscate malicious code…”
Other
1 techniqueRecent activity
26 sources tracked across advisories, community write-ups, and news. New activity surfaces here as Mallory finds it.
Cyber-oriented LLM marketed on underground forums for offensive use without content filtering; described as useful for phishing content or simple malware stubs.
An illicit large language model variant referenced as being used in scam operations to improve social-engineering content and remove linguistic indicators that might expose operators.
Commercially available malicious LLM used to support phishing, fraud, malware development, and attack automation in the cybercrime ecosystem.
AI-based hacking tool distributed via underground and open platforms; described as enabling phishing automation, malware development, reconnaissance, brute force, vulnerability exploitation, and social engineering.
The version that knows your environment.
Match every observed IP, domain, and hash against your live telemetry.
Named campaigns wielding this family, with evidence pinned to each claim.
CVEs this family uses for access and lateral movement.
YARA, Sigma, Snort, and vendor rules, auto-deployed to your SIEM.
Every documented technique, ranked by evidence weight.
Reddit, Mastodon, and CTI community discussion around this family.