Now that I’ve passed these exams, I wanted to share some thoughts from a security perspective, along with some general observations:
AI1-C01: AWS Certified AI Practitioner (Beta) – Passed
I was pleasantly surprised to see that security topics make up about 15% of the exam.
– Covers general AWS security (IAM, CloudWatch, CloudTrail)
– Focuses on Bedrock security
– Guardrails are fantastic for ensuring the intended focus of LLMs and preventing harmful responses.
– Delves into Amazon Q security
– Explores the different types of problematic prompts
– Discusses data security in general, including methods for removing PHI and PII
– Addresses privacy concerns and the potential abuse of prompts, as well as how the data you provide enhances foundational models
– Emphasizes interpretability and explainability in AI responses
– Introduces model cards

ME1-C01: AWS Certified Machine Learning Engineer – Associate (Beta) – Passed
Again, security is comprised of about 15% of the exam.
– Builds upon the AI topics mentioned earlier
– Highlights that SageMaker has its own security model layered on top of IAM, offering tighter control for ML/Data teams
– Goes more in-depth on compliance and regulatory requirements
– Started learning about ISO 42000 (I’ll dive deeper into that in another post)
– Touches on upcoming regulations in different countries (more on this in a future post)
– Covers VPC security for everything entering and exiting the ML toolset—from data input, training, modeling, and deployment, to data output
– Discusses security models for usage, teams, and sharing
– Focuses on interpretability and explainability of model internals
– Reinforces the importance of model cards

Both of these exams require some exposure to machine learning, regardless of the specific AWS tools. If you don’t have a solid understanding of concepts like underfitting, overfitting, bias, variance, and recall vs. precision, you might struggle with 20% to 35% of the questions (even more so in the Machine Learning cert). Additionally, if you’re unfamiliar with AWS fundamentals like IAM, Security Groups, Auto Scaling, VPC networking, etc., you could find about 25% of the exams challenging. The rest focuses more on using ML and AI tools to overcome obstacles or enhance capabilities using specific AWS ML and AI offerings.

I learned most about the security features available in AWS’s AI and ML services and how to leverage many of the SageMaker and ML tools that make MLOps a better place. It’s surprisingly easy to get started with AI and ML using the tools AWS provides for non-developers, which can then be handed over to data and ML engineers to maximize the potential of pre-made conceptual models.

It’s time to return to general security and finish getting my CISM.

Feel free to message me if anyone is interested in chatting about ML or AI security over coffee, in a small group, or even at a speaking engagement.
#ML #AI #machinelearning #artificialintelligence #aws