This page provides answers to frequently asked questions about Fluid Attacks' API and IDE plugin, especially privacy concerns regarding the features powered by Claude Sonnet.
To begin using the API, we recommend you read the articles on this Knowledge Base's Use the API section. Bear in mind that you will need prior knowledge of the GraphQL language to make requests to the API.
Fluid Attacks uses large language models (LLMs), i.e., artificial intelligence designed for advanced text processing and generation. Based on enormous databases, LLMs can create natural language content and even code with accuracy and consistency.
Anthropic's Claude 3.5 Sonnet in Amazon Bedrock.
Claude Sonnet in Fluid Attacks' IDE plugin or extension is crucial in generating code-based remediation guidelines ("Custom fix" functionality) and automatic code correction ("Autofix" functionality). The process begins with extracting a specific code fragment from the selected vulnerable file. This fragment is securely sent to the Claude 3.5 Sonnet instance hosted by Amazon Bedrock, and a response with remediation suggestions is later obtained.
The code sent to Claude Sonnet is interpreted using the context provided at the function/class level, specifically about the line of code containing the vulnerability. This AI model has no global knowledge of the source code or the business logic of the application under evaluation. It is clarified that its access to the code is limited to a small piece representing a specific function.
The Claude 3.5 Sonnet model is hosted by Amazon Bedrock, which does not retain data.
Fluid Attacks understands the importance of maintaining the confidentiality and security of its customers' code. Fluid Attacks ensures compliance with strict privacy and data security policies when employing AI for vulnerability management. The key policies in Amazon Bedrock, which hosts the Claude 3.5 Sonnet model used by Fluid Attacks, can be expressed as follows:
Fluid Attacks' hacking team uses this tool in its daily work in vulnerability reporting.
Fluid Attacks' IDE extension uses or considers all the available information of the repositories listed in the platform's Scope section.
Fluid Attacks uses the Claude 3.5 Sonnet model.
No. Fluid Attacks' initiative started with Visual Studio Code (VS Code), which is recognized as one of the industry's most widely used integrated development environments (IDE). This choice is supported by its outstanding popularity, extensibility, and robustness within the development community.
Although it is technically possible to test from the IDE —as several automated tools do— it often lacks security rigor. For example, security testing could be outside the control of management and security teams and follow the developers' discretionary frequency.
On VS Code, select View > Command Palette... and then type the command Fluid Attacks: Set Fluid Attacks token. When you select the command, a box appears where you can paste the token.