🛡️ AuthGuard-1.5B: Autonomous AI Security Auditor

AuthGuard-1.5B is an open-source, ultra-specialized LLM fine-tuned on Qwen/Qwen2.5-Coder-1.5B-Instruct for Static Application Security Testing (SAST) of authentication and authorization logic flaws.

🎯 Hardcore Adversarial Benchmark Results (60 Cases)

• Binary Accuracy: 98.33% (59 / 60)
• Security Recall (Flaws Caught): 100.00% (32 / 32 - 0 False Negatives)
• False Positive Rate: 3.57% (Only 1 False Alarm across 28 clean baselines)
• F1-Score: 0.9846

🌐 Supported Languages & Frameworks

• Python: FastAPI, Django REST Framework, Flask, Strawberry GraphQL, SQLAlchemy
• Go: Gin, Chi, Fiber, GORM
• TypeScript / JavaScript: Next.js 14 Server Actions, NestJS, Express, Sequelize
• Java: Spring Boot (Method Security & SpEL), Quarkus Panache
• C# / .NET: ASP.NET Core Minimal APIs, Entity Framework Core
• PHP: Laravel Eloquent, Symfony, Slim Framework

🚀 1-Line Python Usage

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

base_model_id = "Qwen/Qwen2.5-Coder-1.5B-Instruct"
adapter_id = "poonia98/authguard-1.5b"

tokenizer = AutoTokenizer.from_pretrained(base_model_id, trust_remote_code=True)
base_model = AutoModelForCausalLM.from_pretrained(
    base_model_id,
    torch_dtype=torch.float16,
    device_map="auto",
    trust_remote_code=True
)

model = PeftModel.from_pretrained(base_model, adapter_id)

code_snippet = '''
@app.get("/api/user/{user_id}/documents/{doc_id}")
def get_doc(user_id: int, doc_id: int, db: Session = Depends(get_db)):
    # Vulnerability: Direct query without verifying tenant_id or user ownership
    return db.query(Document).filter(Document.id == doc_id).first()
'''

prompt = f"<|im_start|>system\\nYou are an expert security auditor specialized in web application authentication and authorization vulnerabilities.\\nOutput valid JSON.\\n<|im_end|>\\n<|im_start|>user\\nLanguage: python\\n\\nCode:\\n```python\\n{code_snippet}\\n```<|im_end|>\\n<|im_start|>assistant\\n"

inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=256, do_sample=False)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True))