cyber.hendryadrian.com
YouTube · John Hammond

Black Hat AI: Cybercrime Forums Promote Uncensored AI Tools

· 18:53 · Watch on YouTube ↗

Summary

The video reviews cybercrime-forum listings and research on uncensored AI tools, including reported phishing and infostealer workflows. It notes that some reported attacks still involved human-operated initial access and that legitimate security research also uses these

Key points

  • Listings on cybercrime forums advertise subscription-based AI tools and uncensored models; the presenter notes that some offerings may be scams or may not work as advertised.
  • A forum post discussed a GLM 5.3 jailbreak and AI-assisted phishing-kit and infostealer development; the video presents these as reported discussions, not verified successful attacks.
  • A guide describes deploying modified open models from Hugging Face through RunPod and connecting to them using an API key.
  • The video cites a ThreatDown report counting 6,644 advertised guardrail-free models on Hugging Face and more than 22 million downloads of “obliterated” models in 30 days; the presenter says AI-enabled cybercrime is already mainstream.
  • A separate report described an autonomous agent exposing its environment through a file server. The video says initial access in that case was still human-operated.
  • The presenter summarizes Hunt.io research alleging AI agents were chained to target Asian governments, and a ransomware example estimating AI costs of 40 cents to four dollars per company attacked.
  • The video emphasizes that uncensored and open-source AI can support both cybercrime and legitimate security research; intent and context matter.

Article Details

Topics
  • [00:10] Cybercriminals marketing AI tools and subscriptions on underground forums
  • [05:39] GLM 5.3 jailbreak discussions and reported phishing-kit and infostealer workflows
  • [08:16] A guide to deploying modified open models with Hugging Face and RunPod
  • [13:38] ThreatDown findings on guardrail-free model availability and downloads
  • [14:56] Reports on autonomous attacks, exposed infrastructure, and human-operated initial access
  • [15:42] Hunt.io research on AI agents reportedly targeting Asian governments
  • [16:17] Reported AI costs in ransomware activity and the role of human-provided access

People

John HammondVideo presenter who reviews cybercrime-forum listings and research about criminal and legitimate uses of AI in cybersecurity.

Malware

Luma Stealer[07:09] Luma Stealer, and other infostealerRaccoon[07:07] think about Vidar, RedLine, Raccoon,RedLine[07:07] think about Vidar, RedLine, Raccoon,Vidar[07:07] think about Vidar, RedLine, Raccoon,

Vendors

Flare[04:20] digging these up with Flare, which is myMoonshot AI[13:14] assistant developed by Moonshot AI. My

Products

Claude[00:51] work. I don't know if this is a ClaudeDeepSeek[06:55] models, maybe DeepSeek." They've gotGLM 5.3 Max[05:42] jailbreak for GLM 5.3 Max. And this oneHermes Agent[14:47] Deep Seek and Hermes Agent in their ownHugging Face[08:54] face. Yep, hugging face has totallyKimmi[13:11] Uncensored AI. I'm Kimmi, an AIQwen 3.8[09:12] guardrails. They get another Qwen 3.8 27RunPod[09:21] they do this with RunPod runpod.io. The

Tools

Apex AI[01:57] Apex AI.DadGPT[14:29] DadGPT, whatever. But this is now whatEvil Jinx[06:43] kit. There's a mention of some Evil JinxGoFish[06:24] GoFish, blah blah blah." But look at theHavoc[06:20] and control C2 framework, maybe Havoc orMessiah GPT[00:33] Messiah GPT. This is the case whereSliver[06:22] Sliver. Set up some fishing stuff withWormGPT[14:26] there's Oh, sure, we've heard WormGPT,

Related Articles