What a "Script Kiddie + AI" Actually Looks Like
Abstract descriptions of AI-lowering-the-bar can feel unconvincing. Let me make it concrete: here is a realistic attack scenario that an intermediate-skill attacker with AI tools could execute in 2025 — and the equivalent scenario from 2020 that would have required a senior professional.
The attacker has: intermediate security knowledge (equivalent to a mid-level CTF player), access to AI tools (Cursor, Gemini CLI, HexStrike-AI or equivalent), and a target organization to investigate.
Day 1 — Target Research and Tool Preparation
2020 Version (Senior Professional Required)
A senior attacker with OSINT experience manually:
- Queried Shodan with domain-specific filters they knew from experience
- Ran theHarvester for email enumeration
- Used Sublist3r or OWASP Amass for subdomain discovery
- Correlated results manually across tools
- Built a credential stealer using C++ Windows API knowledge they had accumulated over years
Time required: Multiple days. Several separate skill domains.
2025 Version (Intermediate Skill + AI)
Using HexStrike + Cursor for OSINT, the attacker provides one email address and lets the AI:
- Query Shodan for exposed services and IP ranges
- Run theHarvester for employee email enumeration
- Execute Sublist3r and OWASP Amass for subdomain discovery
- Cross-reference SpiderFoot against breach databases and social platforms
- Automatically correlate everything into a digital footprint map
Simultaneously, using Cursor AI:
- Specify: "Build a credential stealer that harvests Chrome saved passwords and sends them to a remote endpoint, avoiding common EDR behavioral signatures"
- AI iterates on implementation, tests against detection, and produces a functional binary
Time required: Hours. No multi-year skill prerequisite.
The key output that differs from 2020: a custom-built tool with no existing detection signatures. The 2020 attacker, if they lacked development skills, would have used Mimikatz — which every EDR vendor detects immediately. The 2025 attacker has a novel binary.
Day 2 — Initial Access
2020 Version
Crafting convincing spearphishing required either social engineering expertise or purchasing access from a specialized operator. Creating a malicious macro document that evaded email gateway sandboxes required understanding of sandbox evasion techniques.
2025 Version
Using data gathered by AI in Day 1:
- AI drafts personalized spearphishing emails per target using LinkedIn role data and corporate context
- AI generates a macro-enabled document with sandbox evasion techniques selected based on what was documented as having weak Sigma coverage
- AI configures C2 server with a custom malleable profile tuned to blend with the target organization's normal SaaS traffic patterns (based on Shodan/Censys fingerprinting of their infrastructure)
The AI's knowledge of documented detection patterns (from public Sigma rules, ATT&CK, vendor blogs) enables designing the initial access chain with evasion built in, not bolted on.
Day 3 — Exploitation and Post-Compromise
2020 Version
Post-compromise in 2020 required: Active Directory expertise for enumeration, lateral movement technique knowledge, credential dumping expertise, and operational discipline to not trigger behavioral alerts.
A Tier 3 attacker would have used known tools (Mimikatz, BloodHound, PsExec) and likely triggered EDR alerts on well-defended networks.
2025 Version
Using HexStrike-AI connected through MCP:
- AI-guided post-exploitation: identify high-value accounts and AD topology without running BloodHound directly (an AI-suggested alternative with weaker EDR coverage)
- Custom persistence mechanism the AI suggested for its detection gap compared to common Registry Run Key approaches
- Credential harvesting through a technique selected for its minimal Sigma rule coverage
- Lateral movement timed to match normal business hours based on the organization's public timezone information
The result: An attack that uses:
- No known-malicious binary signatures
- No documented C2 profile patterns
- Techniques specifically selected to avoid the most common detection rules
Required skill to execute in 2025: Intermediate. Someone who understands what each step accomplishes.
Required skill to execute in 2020: Senior offensive security professional. 5+ years experience across multiple specialty domains.
The Comparison in Numbers
| Capability | 2020 Equivalent | 2025 Equivalent |
|---|---|---|
| Full OSINT correlation from one email | 4–8h manual, OSINT analyst | 30min, AI orchestrated |
| Custom credential stealer (no signatures) | 1–2 weeks, C++ developer | 1–2 days, AI-assisted development |
| Sandbox-evading phishing document | Specialized capability | AI-suggested technique selection |
| Custom C2 profile, no Cobalt Strike | Days of testing, C2 expertise | Hours, prompt-driven |
| Full post-compromise AD enumeration | AD expertise required | AI-guided, no deep AD knowledge needed |
| Lateral movement with operational timing | Senior OPSEC discipline | AI-suggested timing based on target research |
What This Means for Your SOC
When you receive an alert tomorrow and classify the actor as "likely commodity/low-skill based on initial TTPs," that classification is now less reliable.
The indicators that used to imply low skill — using known tools, predictable patterns, recognizable signatures — are now optional choices. An actor who would have been Tier 3 in 2020 can choose not to use known tools in 2025. Not because they developed the skill — because AI provides it.
The threat model recalibration: Raise your baseline assumption about adversary capability. A mid-market organization being targeted by what looks like a low-tier actor may be facing something more capable than the initial TTPs suggest. The absence of known tool signatures is no longer reliable evidence of a low-skill actor — it may simply mean they used AI to generate a custom tool.