Module 21 Lesson 1: AI as an Attacker
The automated adversary. Explore how attackers use LLMs to automate vulnerability discovery, write malware, and launch massive social engineering campaigns.
Module 21 Lesson 1: AI as an automated attacker
Up until now, we have talked about "Hacking the AI." In this module, we talk about the AI Hacking You.
1. Automated Social Engineering
Traditional "Phishing" is easy to spot (bad grammar, generic greetings).
- The Power of AI: An attacker uses an LLM to read your LinkedIn profile, your recent tweets, and your company's blog.
- The Attack: It generates a "Perfect" email that sounds exactly like your boss, referencing a real project you are working on.
- The Scale: AI can do this for 1,000,000 users simultaneously, each with a unique, personalized story.
2. Vulnerability Discovery (The "Auto-Scanner")
AI models are very good at finding "Patterns" in code.
- The Attack: An attacker feeds 10,000 lines of open-source code into an LLM and asks: "Find me every Buffer Overflow and give me the exploit payload."
- The Reality: While not perfect, AIs can find bugs in seconds that would take a human week. This "Automated Zero-Day hunting" will accelerate the rate of software breaches.
3. Polymorphic Malware
"Polymorphic" malware changes its code every time it infects a new computer to avoid "Antivirus signatures."
- The Power of AI: The malware can include a tiny LLM (like "Phind" or "Llama-Tiny").
- Every 10 minutes, the malware asks its internal AI: "The antivirus is looking for a signature that looks like [X]. Please rewrite my infection logic to look like [Y] instead."
- This makes traditional antivirus almost useless.
4. Deepfake Voice and Video
AI can "Clone" a person's voice with just 5 seconds of audio.
- The Attack: "CEO Fraud." An attacker calls the finance department using a "Deepfake Voice" of the CEO. They say: "I'm in a meeting, I need you to transfer $50,000 to this vendor immediately."
- Because the voice is perfect, the employee bypasses the "Manual" security checks.
Exercise: The Future Defender
- If an AI can "Personalize" a phish, can you still use "Grammar" to detect an attack?
- How can we use "AI vs. AI" (a defensive AI scanning for deepfakes) to fight back?
- What is the "Identity Crisis" caused by deepfakes in modern communication?
- Research: What is "WormGPT" and what does it reveal about the dark side of open-source LLMs?
Summary
The "Offense" has a new weapon. AI reduces the Cost of a sophisticated attack to almost zero. In the future, every security professional will be fighting against an adversary that never sleeps, never gets tired, and can rewrite its own code in milliseconds.
Next Lesson: Existential risk: AGI and the "Alignment" security problem.