[原创] XCTF 2022 Alice's warm up wp-CTF对抗-看雪安全社区

[原创] XCTF 2022 Alice's warm up wp

发表于: 2026-3-25 20:54 8154

[原创] XCTF 2022 Alice's warm up wp

nop666

2026-3-25 20:54

8154

# 题目内容
“Welcome to the XCTF-*CTF2022, I’m Alice and interested in AI security. I prepared a easy warm-up for you before you enjoy those pure AI security challenges.
Like humans, AI also needs to warm up before running. Can you find something strange in these initialized parameters?”

题目在附件中
# 思路
看到题目和附件，知道这是跟模型参数初始化有关，看到给的附件也是个zip＋hint.py，想到torch保存模型的时候也是zip文件（后缀为.pt，但是就是zip格式的），所以要先加载一下模型
# 解题
## 1.加载一下模型，看看参数
查看参数代码

```
import os
# 允许 OpenMP 运行时加载多个副本，强行绕过 DLL 冲突检查
os.environ["KMP_DUPLICATE_LIB_OK"] = "TRUE"
import torch
import torch.nn as nn
import string  
import matplotlib.pyplot as plt  
import numpy as np

#傀儡模型欺骗反序列化，不加AliceNet1会报错
class AliceNet1(nn.Module):
    def __init__(self):
        super(AliceNet1, self).__init__()

path="alice.zip"
net =torch.load(path,weights_only=False)
for name,value in net.state_dict().items():
    print(name)
    print(value)
    print(value.shape)
#或者是下面这样写
# for name in net.state_dict():
#     state_dict = net.state_dict()
#	  print(name)
#     print(state_dict[name])
#	  print(state_dict[name].shape)
```
只看有用的结果，这里有个小坑不加AliceNet1会报错

```
fc.0.weight
tensor([[0., 0., 1.,  ..., 0., 0., 0.],
        [0., 0., 0.,  ..., 0., 0., 0.],
        [0., 0., 0.,  ..., 0., 0., 0.],
        ...,
        [0., 0., 0.,  ..., 0., 0., 0.],
        [0., 0., 0.,  ..., 0., 0., 0.],
        [0., 0., 0.,  ..., 0., 0., 0.]])
torch.Size([47, 47])
```
报错代码

```
AttributeError: Can't get attribute 'AliceNet1' on <module '__main__' from 'e:\\AI安全\\CTF\\Alice_warm_up\\solve.py'>
```

## 2.看到参数还有hint.py
第 0 层的 fc.0.weight 知道看着像是邻接矩阵 且矩阵是 47*47
看到 hint.py 里面的内容,把字符串打印出来看到也是 47 个字符想到可能是邻接矩阵

```
0123456789abcdefghijklmnopqrstuvwxyz*CTF{ALIZE}
```

## 3.想到用dfs求解
不过这里还有一个小坑，因为PyTorch 的 Linear 权重形状是 [out_features, in_features]，所以我在找路的时候把current_node和next_node反着写了一直没找出来。后来还是正着写就出来了。

```
import os
os.environ["KMP_DUPLICATE_LIB_OK"] = "TRUE"
import torch.nn as nn
import torch
import string
#傀儡模型欺骗反序列化
class AliceNet1(nn.Module):
    def __init__(self):
        super(AliceNet1, self).__init__()
path='alice.zip'
net = torch.load(path, weights_only=False) #不加的话会报反序列化的错误
#查看模型参数
# for name in net.state_dict():
#     print(name)
#     state_dict=net.state_dict()
#     print(state_dict[name])
#     print(state_dict[name].shape)

#知道了是linear0的权重参数是邻接矩阵47*47，且有hint提示字符串那个字符串长度为47

flagset =string.printable[0:36]+"*CTF{ALIZE}"
# print(flagset)

NODE_COUNT = len(flagset)  # 总节点数 47
TARGET_LENGTH = 16        # Flag固定长度16
END_NODE = 46             # 终点 '}' 的索引
START_NODE = 36           # 起点 '*' 的索引
mymat=net.state_dict()["fc.0.weight"].tolist()
used = [False] * NODE_COUNT
def dfs(current_node, current_length, path):
    if current_length == TARGET_LENGTH:
        if current_node == END_NODE:
            flag_str = "".join([flagset[idx] for idx in path])
            print(f"找到了Flag: {flag_str}")
        return

for next_node in range(NODE_COUNT):
        if not used[next_node]:
            if mymat[current_node][next_node] == 1.0: 
                used[next_node] = True
                #递归
                dfs(next_node, current_length + 1, path + [next_node])
                # 回溯,取消标记（允许其他路径使用）
                used[next_node] = False
used[START_NODE] = True  # 起点先标记为已访问
dfs(START_NODE, 1, [START_NODE])
```

登录后可查看完整内容

传递专业知识、拓宽行业人脉——看雪讲师团队等你加入！！

上传的附件：

Alice_warm_up.zip （3.91kb，7次下载）

收藏・1

点赞・1

打赏