HoundDog.ai static code scanner not only flags PII leaks in plaintext within logs, files, cookies, and tokens but also tracks data flows to third-party integrations, highlighting data processing agreement violations before they become production issues.
Hey Amjad,
How does HoundDog handle false positives, especially in complex codebases?
Do you have any plans to integrate with popular CI/CD pipelines for automated scanning?
Congrats on the launch!
Thank you and great questions!
At HoundDog.ai, we use both traditional code analysis methods and LLMs. I'd like to briefly explain the drawbacks of each approach that result in false positives, and how we've addressed them.
We parse abstract syntax trees and match code segments against known regex patterns - a traditional approach still effective in many static code scanners today. Crafting these patterns can be challenging: overly restrictive patterns miss real issues, while broad ones generate false positives. Fortunately, after extensive testing we were able to come up with clever tricks and schemas that allow us to write our patterns systematically while drastically minimizing false positives. With near 100% unit test coverage, our iteration process at this point is quite robust and reliable. We're also leveraging LLMs in moderation to continuously refine our patterns, ensuring they evolve without regressions.
LLMs are getting incredibly good at extracting meaning from code, and we believe they are the future of code analysis. But as of today, due to context length limits and frequent inaccuracies, you cannot throw millions of lines of code at AI and expect results that are consistent, fast, and correct all at the same time. While the inference cost is getting cheaper, expenses can add up quickly if you have to pay for an API or GPU cluster to process repositories with millions of lines of code. Some of the larger businesses we've talked to are not comfortable with giving AI their code or are not ready due to lack of proper legal and engineering infrastructure (although this will surely change in the future). This is why we are currently limiting the usage of LLMs to very specific use cases. One such use case is enhancing our regex patterns as mentioned above. And we extract only a small subset of tokens that we consider critical before sending it over to AI for verification. This is one of the reasons why our scanner can be blazingly fast. In fact, we can confidently claim that it is one of the fastest in the market. So please feel to give it a try!
And yes, we support all popular CI/CD pipelines such as GitHub Actions, GitLab CICD, Jenkins etc.
Report
Congrats ro the launch! 🚀
Report
Hi Amjad, congratulations on the launch! Excited to see how HoundDog.ai will empower teams to integrate security seamlessly into their development processes.
Congrats to the HoundDog.ai team on the launch! This tool sounds like a must-have for identifying and preventing PII leaks. Does it also provide recommendations for fixing flagged issues or just highlight them?
Report
@vietpham yes right now it provides a remediation strategy for each flagged issue
Report
Congratulations @amjad_afanah1 ! Does it provide any specific reports or insights on how to fix the flagged PII leaks and data processing agreement violations?
@zishaniqbal Indeed. We provide detailed remediation strategies for each issue we identify. The good news is that all these PII leak issues can be resolved by either (a) omitting the data altogether (e.g., if you're exposing authentication tokens or social security numbers in a log, they may not be necessary at all), or (b) sanitizing the data through methods like masking, encryption, or tokenization, if needed.
Outcome
Seamailer
HoundDog.ai Static Code Scanner
Telebugs
HoundDog.ai Static Code Scanner
Diaflow
HoundDog.ai Static Code Scanner