With ChatGPT being all the rage, I decided to see if she (it?) could write my next blog post for me. I’d already written all the Steampipe queries and determined the security value behind the blog post. I just didn’t feel like writing it.
The post’s topic was on enumerating your network-plane cloud perimeter.
I started with a simple prompt:
Let’s define the customer’s cloud network perimeter as all the resources that are accessible via the public internet that do not require authentication using cloud provider API access keys. This would include resources that are managed by the cloud provider and exist on the internet along with resources that are managed by the customer, have a public IP address and exist in the customer’s VPC.
I then went on to prompt it with the flow I wanted the post to have. I documented the prompts and direct responses in Google Docs, which you can see here.
The first draft I submitted cleaned up the prompts and removed some extraneous comments that didn’t flow (like comments about firewalls and VPNs). That draft is here, and clocks in at 1722 works and six pages.
Through internal reviews and copy-editing, almost all of the ChatGPT language was deemed either 1) irrelevant to the point of the post or 2) so basic in nature that the audience didn’t need it (seriously, we all know what an IP Address and DNS is). I think the only part I kept was defining how an API Gateway URL was structured (which, ironically, was given to me in an answer about Lambda Config URLs). We also went through a number of rounds on branding, messaging, etc., and cloud-network perimeter became cloud attack surface. Clearly, I’ve been in the presence of AWS PR & Messaging.
The final version is about 1017 words; over 40% of the content was removed. About all ChatGPT did for me was serve as a rubber duck I could talk to when figuring out how to layout this post.