Fortifying AI Against Threats

My name is Vinny, and I am an international student from São Paulo, Brazil. I am currently a sophomore at Whitman College, pursuing a major in Computer Science. This summer of 2024, I have the incredible opportunity to intern at HiddenLayer, a startup based in Austin, TX, focused on protecting AI from adversarial attacks. HiddenLayer excels at unifying two of the biggest assets in the Computer Science world: cybersecurity and artificial intelligence, making this internship exceptionally engaging.

HiddenLayer operates in a unique manner, especially when compared to pre-COVID times. It is a fully remote company, with team members contributing from around the world. I am based in Lynnwood, a city near Seattle, Washington. Initially, I was apprehensive about working in a fully remote environment, but the experience has been impressive even before the internship began. HiddenLayer sent me a laptop, along with some very cool swag, including shirts, sweaters, pants, backpacks, and pens. On the first day, there was a meeting with all the interns, and it was great to see everyone wearing the company swag. Following that, the Director of Engineering added us to the Engineering channel on Slack, where we received a warm welcome from the team, making remote work much more enjoyable than I had anticipated.

At HiddenLayer, the product includes a user interface (UI) where consumers can upload their machine learning models to be scanned by our Model Scanner. The Model Scanner detects if the model contains any malicious code or vulnerabilities, such as susceptibility for prompt injections in large language models. Ensuring the UI is fully functional and interactive is crucial. Within the Engineering team, there is a dedicated testing team, and within that team, a UI testing team, where I am currently working.

One might think testing an UI is simple: press buttons, scan models, assess outputs, and if it works, great. However, the process is much more complex. For instance, we need to account for scenarios where servers running the product may fail on AWS, or the website might be down when no one is testing. We must ensure it is operational 100% of the time, which is where automated testing comes in. At HiddenLayer, we use a powerful tool developed by Microsoft called Playwright. This tool, combined with a programming language of your choice, allows us to automate end-to-end (e2e) activities on the website across various browsers like Microsoft Edge, Google Chrome, and Safari. My role at HiddenLayer involves writing e2e tests for the UI using Playwright and refactoring existing tests to improve coverage and resilience.

Throughout this journey of writing automated tests, I have learned an immense amount. Understanding why a test fails requires a deep comprehension of how the website and the APIs work. For debugging, we utilize various tools, such as the developer tools in Chrome and the reports generated by Playwright after a test fails. Fixing these issues is not always straightforward. If an automated test that runs multiple times a day fails, we must either fix it or contact the responsible person for the change that caused the break if that is the case.

This internship at HiddenLayer is immensely important for my career. It not only provides me with practical experience in cutting-edge technologies but also enhances my skills in a real-world, professional environment. The knowledge and experience I am gaining here will be invaluable as I continue my journey in the field of Computer Science, paving the way for future opportunities and career growth.