some problem about DeepExploit about machine_learning_security

[mbhatt1]

You meant to comment here? Or a different repo?

[13o-bbr-bbq]

i received this issue from anonymous using e-mail.
so posted this issue at this instead of anonymous.

[opsoyo]

I am having a similar experience in Kali after training.

[*] Port scanning: ###.###.###.###
[*] Executing keep_alive..
[!] type:<class 'AttributeError'>
[!] args:("'NoneType' object has no attribute 'decode'",)
[!] 'NoneType' object has no attribute 'decode'
[!] Failed: version

[hustzhm]

Hi,
I have some troubles about the DeepExploit.

1. For the same target server,the trained model is different.I do not change the parameters about the network,and the target server is metasploitable2-Linux,referred in your GitHub. I want to know how to solve the problem because the different models make the exploit unstable,which can not meet our demands.Perhaps,does the instability of RL lead to the problem?
2. Every time I use the tool to exploit the same target server,I will get different results.The model sorts the action(payload) by the probability.According to common sense,the same payload should be selected every time for the same port when the model is trained.But the situation is not like that.
3. I can not use the tool to exploit the other servers.It always failed to exploit.So I just train and test the model in the same server.
   Test result .docx

[rongqinglee]

I have tried your DeepExploit system and finished the configuration.
After training, I used the system to exploit one machine, it show some failed message below.

[!] type:<class 'AttributeError'>
[!] args:("'int' object has no attribute 'decode'",)
[!] 'int' object has no attribute 'decode'
[!] Failed: module.info

Is there something i missed or ?

I encountered this problem too. Some version of Kali gives rank as 600, 500, ..., but not 'excellent', 'great',...

I modify some code as following and could work:
#rank = module_info[b'rank'].decode('utf-8')
#if rank in {'excellent', 'great', 'good'}:

modify as ==>

rank = module_info[b'rank']
if rank in {600, 500, 400}:

[NovaBlastHD]

Hi,
I have some troubles about the DeepExploit.

1. For the same target server,the trained model is different.I do not change the parameters about the network,and the target server is metasploitable2-Linux,referred in your GitHub. I want to know how to solve the problem because the different models make the exploit unstable,which can not meet our demands.Perhaps,does the instability of RL lead to the problem?

2. Every time I use the tool to exploit the same target server,I will get different results.The model sorts the action(payload) by the probability.According to common sense,the same payload should be selected every time for the same port when the model is trained.But the situation is not like that.

3. I can not use the tool to exploit the other servers.It always failed to exploit.So I just train and test the model in the same server.
   [Test result .docx](https://github.com/13o-bbr-bbq/machine_learning_security/files/3381708/Test.result.docx)

To answer your first two questions, which are pretty much the same thing, you have to understand how machine learning works. The weights are initialized randomly. I'm not sure if that is the case here but it seems like it. Because they are initialized randomly, when the weights are updated they probably fall inside a local minimum. This is why you get different results each time. If you were to initialized the weights with 0 every time then the model wouldn't be as effective and you would end up with the same data each time.

[1939552724]

@hustzhm
Hi, I try to run this project! But there are some problems and bugs that I can't solve! Could you show me your sucessful project?

one problem like this:

[] 311/2006 Loaded exploit: linux/samba/trans2open
[!] 312/2006 linux/smtp/apache_james_exec module is danger (rank: normal). Can't load.
[] 313/2006 Loaded exploit: linux/smtp/exim4_dovecot_exec
[*] 314/2006 Loaded exploit: linux/smtp/exim_gethostbyname_bof
[!] type:<class 'KeyError'>
[!] args:(b'rank',)
[!] b'rank'
[!] Failed: module.info