Black Hat AI

Black Hat AI

by John Hammond · View the channel on PSI

Formal & Physical Sciences Computing & Cybersecurity URComputer securityURHComputer fraud and hacking
🎙 John Hammond 👥 2.2M 📅 October 1, 2026 ⏱ 18 min 👁 130K 📄 news review 🧭 2026-10-15
Available in: English (current) Français

Keywords

black hat AIdark webthreat actorsjailbreaksubscription model

Summary

John Hammond explores how cybercriminals are actually using AI by browsing dark web forums revealed via Flare. He shows several commercial offers: Messiah GPT ($10/month) and Apex AI ($100/month) which wrap uncensored models, though scammy. He discusses a GLM 5.3 jailbreak and a guide for deploying abliterated models on RunPod using open-source models like Qwen. Threat actors share tutorials for phishing kits and infostealers. He highlights that AI-enabled attacks are commoditized, with costs as low as $4 per attack. He references a ThreatDown report showing 6,644 guardrail-free models on Hugging Face with 22 million downloads. Also mentions Unit 42 and Hunt.io cases of autonomous attacks. Concludes that AI has lowered the barrier for cybercrime, but good-faith security research also benefits, emphasizing the need for vigilance.

125 words

Critical Evaluation

Value of the Information & Strength of the Argument

The video provides tangible evidence from dark web forums, including screenshots and prices, reinforcing the argument that AI for cybercrime is commoditized. John argues effectively that the threat is real but nuanced, not Skynet-like doom. He balances hype with concrete examples and industry reports, though some commentary is humorous.

Scientific Rigor, Source Quality, Title Accuracy

The host cites multiple external sources: Unit 42’s report on autonomous AI attacks, Hunt.io’s analysis of Chinese operators using Claude/Qwen/DeepSeek, and Cybernews on ransomware costs. These are reputable cybersecurity outlets. The title accurately reflects the content. Public comments are overwhelmingly positive, praising the informative content and humor.

111 words

Title / Content Match

Title accurately reflects the content: a survey of black hat AI offerings and tactics on the dark web.

Quality & Reliability

8/10

The video presents real dark web listings and references multiple credible cybersecurity reports (Unit 42, Hunt.io, Cybernews), balancing technical demonstration with industry data. Some content is anecdotal but overall reliable.

Key Moments

Markers derived by PSI from the transcript: the creator did not define chapters.

Cited Sources

Concurring Sources

  • Unit 42 report — Independent research corroborating AI autonomous attacks
  • Hunt.io blog — Detailed analysis of multi-agent AI hacking campaigns

External References

Contribution & Novelties

This video provides a hands-on, visual tour of real dark web AI offerings, showing actual screenshots and price tags. It demystifies the hype by presenting concrete examples and contrasts with industry reports. The novelty lies in the direct observation of criminal marketplaces and the practical demonstration of deploying abliterated models.

Pour aller plus loin :

  • Abliteration technique on Hugging Face — Explains how refusal vectors are removed from models.
  • RunPod — Serverless GPU platform used to host uncensored models.
  • Qwen 3 model family — Open-source models commonly abliterated.
  • United Nations cybercrime treaty — International efforts against cybercrime.

97 words

Radar Profile

The radar profile shows high scores in information quantity, quality, and reliability, with a slightly lower technical level given the accessible presentation. Strong on evidence and sources, making it a valuable overview of black hat AI trends.

Reliability 8/10

💬 Positive overall. Sur les 30 commentaires analysés, l'orientation est très positive, avec des retours enthousiastes sur la transparence du contenu et l'humour, ponctués d'anecdotes personnelles et de remerciements pour les ressources partagées.