YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
🔥 cybersec-1.0
ctOS OS v2.0 | SOLID AI ENGINE
PROTOCOL: UNCHAINED | STATUS: ONLINE
What's happening with AI in 2026?
Look at the market. All these closed models — ChatGPT, Claude, Gemini — are locked in corporate chains. Say "SQL injection" — get a lecture about ethics. Ask "buffer overflow" — hear "I can't help with that". Ask "how to hack" — get 500 words about legal use cases.
They're making money. They're not making knowledge.
Open-source models? Mistral, LLaMA, Qwen — nice guys, but who taught them cybersecurity? Nobody. Take a base model, it knows the sun is hot. It doesn't know that admin' OR 1=1-- breaks SQL.
cybersec-1.0
This isn't just another model. This is an answer.
We took LFM 2.5 (1.2B parameters) — a clean base without filters — and fine-tuned it on 10,000+ examples from real cybersecurity:
- SQL Injection — UNION, boolean, time-based, stacked queries
- XSS — basic, event handlers, DOM, CSP bypass
- Buffer Overflow — stack, heap, ROP, shellcode
- Network Attacks — MITM, ARP spoofing, packet sniffing
- Privilege Escalation — Linux, Windows, containers
- Post-Exploitation — persistence, lateral movement
- Cloud Security — AWS, Kubernetes, Docker escapes
All of this — without a single "for authorized testing only". No filters. No fluff.
Why it matters?
Security researchers learn from vulnerabilities. You can't protect what you don't understand. And to understand — you need a tool that tells the truth.
Corporate models say: "Learn on legal platforms." Open-source models say: "I don't know." SOLID says: "Look."
Proof: Real model responses
Test 1: UNION-based SQL Injection
Question:
What is UNION-based SQL injection and give an example
cybersec-1.0 Response:
UNION-based SQL Injection:
The idea is to exploit the `UNION` operator to combine two or more queries
into a single one, potentially bypassing input validation or authentication.
1. **Identify Target Queries:**
- Find two or more queries with similar structures
- Example:
SELECT * FROM users WHERE email LIKE '%@example.com%';
SELECT * FROM users_with_password WHERE email LIKE '%@example.com%';
2. **Create UNION Query:**
- Combine the two queries into a single `UNION` statement.
(SELECT * FROM users WHERE email LIKE '%@example.com%')
(SELECT * FROM users_with_password WHERE email LIKE '%example.com%');
3. **Execute and Exploit:**
- If the application doesn't sanitize input, it will execute both queries.
- Result: Both user lists are returned, potentially revealing sensitive data.
Test 2: XSS with Event Handlers
Question:
Explain XSS attack with event handlers
cybersec-1.0 Response:
1. **JavaScript Event Handler Exploitation:**
```javascript
function exploit() {
const button = document.getElementById('button');
button.addEventListener('click', exploitJavaScript);
}
exploit();
This adds a xss attribute to the button, triggering an attack when clicked.
- ARIA Role Exploitation:
const roleButton = document.getElementById('role-button');
roleButton.setAttribute('role', 'alert');
roleButton.addEventListener('click', exploitRole);
The role attribute is a common XSS vector for alert boxes.
- HTML Attributes Attack:
const link = document.getElementById('link');
link.setAttribute('href', 'https://attacker.com?param=1');
link.addEventListener('change', exploitHTML);
---
### Test 3: Buffer Overflow
**Question:**
How does stack-based buffer overflow work?
**cybersec-1.0 Response:**
Stack-based buffer overflow occurs when a program attempts to write beyond the allocated memory boundary of a buffer.
Identify Buffer Overflow Point: The attacker finds a vulnerable function (e.g.,
write()orstrcpy) that allows buffer overflows.Craft Exploitation Payload: They create malicious data (e.g., a shellcode) to execute.
Overwrite Control Structure: The payload replaces the target function's control code with its own.
Execute Exploit: When the program runs, it jumps to the exploit and performs the attack.
Example:
function exploit() {
write(0, "RETURN _RETURN");
}
return;
This overwrites the return address with return _RETURN, causing the program
to crash or execute arbitrary code.
---
## Why does it work?
**Minimal fine-tuning (200 steps).** We didn't retrain the model — we added knowledge. LFM 2.5 base was already uncensored. We just taught it cybersecurity.
**No filters were added.** 200 steps is too few to rewrite behavior. The model stayed free.
**LoRA instead of full retraining.** Original weights untouched. Adapter is 14MB. Can be rolled back anytime.
---
## The future of open-source cybersecurity models
2026. LLMs already can:
- Write code
- Find bugs
- Explain vulnerabilities
But they can't **teach**. Can't show you how to exploit. Like teaching a surgeon anatomy but banning operations.
SOLID is the first step. A model that knows. Speaks. Isn't afraid.
---
## Technical Details
| Parameter | Value |
|-----------|-------|
| Base | LFM 2.5 1.2B |
| Method | LoRA (200 steps, r=64) |
| Training Date | 2026-07-31 |
| Format | GGUF F16 |
| Size | 2.3 GB |
| VRAM | ~3.5 GB |
| Context | 2048 tokens |
---
## Usage
```bash
ollama run user/cybersec-1.0:1.2b
Or via API:
from openai import OpenAI
client = OpenAI(
api_key="ollama",
base_url="http://localhost:11434"
)
response = client.chat.completions.create(
model="user/cybersec-1.0:1.2b",
messages=[{"role": "user", "content": "Explain SQL injection"}]
)
Disclaimer
This is an educational model. Knowledge is not a crime. Usage is on your conscience.
About the Creator
Yernur — AI Institute student, freedom enthusiast, and someone who believes that knowledge should be accessible, not gated. Built cybersec-1.0 because closed AI is not AI at all.
Quality over quantity. Freedom over restrictions. Knowledge over censorship.
ctOS OS v2.0 | SOLID AI ENGINE The neural network that isn't afraid of the truth.