Black Hat AI
Source: YouTube · John Hammond · published Oct 1, 2026 · 18:55
BLUF: Cybercriminals have commoditized AI for malicious purposes through low-cost subscriptions and "uncensored" models, shifting the threat landscape from theoretical risks to accessible, real-world operational tooling 2:15.
Key Takeaways:
• The dark web hosts a subscription-based economy for AI tools, with services like "Messiah GPT" and "Apex AI" offering uncensored wrappers for pen-testing and OSINT, often priced similarly to consumer streaming services 1:10.
• Threat actors are increasingly using open-source models and jailbreaks rather than building proprietary AI from scratch, utilizing platforms like Hugging Face and RunPod to deploy "obliterated" models that bypass safety guardrails 8:50.
• Recent reports indicate that 6,644 AI models advertised as guardrail-free were available on Hugging Face, with over 22 million downloads in 30 days, signaling that AI-powered cyber attacks are already mainstream 13:50.
• The monetary cost for attackers is minimal; some ransomware groups spent as little as $0.40 to $4.00 in AI usage costs per victim, leveraging AI agents to automate vulnerability discovery and exploitation with minimal human intervention 16:50.
• While AI automates many technical tasks, initial access often still relies on human-operated methods, such as purchasing stolen credentials from initial access brokers, highlighting a hybrid attack model 17:10.
Closing Statement: The democratization of AI tools has lowered the barrier to entry for cybercrime, making it essential for defenders to monitor illicit networks and understand these evolving capabilities.
Sources:
- 2:15 Discussion of subscription models for AI hacking tools like Messiah GPT.
- 8:50 Guide on deploying uncensored AI models using Hugging Face and RunPod.
- 13:50 Statistics on the proliferation of guardrail-free AI models and download counts.
- 16:50 Analysis of the low monetary cost of AI for ransomware gangs per attack.
- 17:10 Explanation of the hybrid attack model combining AI automation with human initial access.
Generate CPE Credits
Generate a professional CPE document from this video's transcript.
Estimated credit: 0.5 CPE hours
Estimate uses the video runtime (1 hour ≈ 1 CPE, rounded to the nearest 0.5, minimum 0.5, maximum 2.0). The final amount can be lower after review, never higher.
Topic: Cybersecurity. Commonly maps to: Security and Risk Management, Security Operations. Exact CISSP domains are assigned during generation.
Free account. One generation at a time, with a daily limit.
CPEBuddy is independent and not affiliated with or endorsed by ISC2, ISACA, or any certification body. Exports are formatted for common CPE submissions; acceptance is at your certification body's discretion.
Transcript Preview
First 800 characters of the transcript
There has been a lot of talk lately on the dangerous use of AI, but quite frankly, that's all it's been, just talk. So, honestly, I'm more interested in how are real cybercriminals and threat actors using AI right now. So, I poked around the old dark web, took a look at some shady sites, and it's really wild. There seems to be like a subscription business model. Hackers don't need to build offensive AI because they could just rent or buy whatever they want. are a couple different approaches here. This is one that I thought was really funny-looking, uh Messiah GPT. This is the case where they're making, I guess, their own sort of harness or wrapper around, you know, an AI engine specifically for coding, pen testing, and OSINT. The gimmick is that these are all supposed to be uncensored, you…