AWS Certified AI Practitioner

A foundational credential for people who work with AI systems rather than build them. AWS assesses it in one 90 minute exam covering machine learning fundamentals, generative AI, foundation models, responsible AI and the governance around all of it.

At a glance
Cost
$100 AWS lists the exam at $100 USD. Prices in other currencies follow the published exchange rates, and there is no separate registration fee. A retake is charged at the full price and AWS enforces a 14 day wait between attempts.
Format
65 questions in 90 minutes, of which 50 are scored and 15 are unscored trial questions that are not marked as such. Question types are multiple choice, multiple response, ordering, matching and case study. Taken at a Pearson VUE test centre or online with a proctor.
Pass mark
700 on a scaled range of 100 to 1,000. AWS does not publish the number of questions this corresponds to, because the scale equates forms of differing difficulty, and section level feedback on the score report is indicative rather than a per domain pass.
Prerequisites
None. AWS suggests up to six months of exposure to AI and machine learning on its platform, and recommends AWS Cloud Practitioner Essentials first for anyone new to the cloud, but neither is enforced at booking.
Renewal
Valid for three years. Recertification is by passing the current version of this exam again, or by earning AWS Certified Machine Learning Engineer Associate, which supersedes it.
Based on
The AIF-C01 exam guide, version 1.4, which lists five content domains with published weightings and a task statement breakdown under each
What it covers

What it covers. The exam is about judgement rather than implementation. AWS states plainly that developing or coding models is out of scope, and the questions instead ask which technique suits a situation, what a result means, and what could go wrong. That makes it unusually well suited to product people, who spend their time on exactly those decisions.

The weight sits on generative AI. Foundation models and their applications carry 52 per cent between them, covering model selection, inference parameters, prompt engineering, Retrieval Augmented Generation, fine tuning and evaluation. Classical machine learning is only a fifth of the exam, and it is there as grounding rather than as the subject.

Roughly a quarter of the marks come from responsible AI, security and governance together. This is the part candidates from a product background tend to find easiest and the part they most often skip in preparation, which is the wrong way round.

AWS service names matter. The exam is vendor specific, and a question will name a situation and expect you to know that Amazon Bedrock hosts foundation models, SageMaker Clarify looks for bias and Guardrails filters output. Learning the concept without the service that implements it will not be enough.

The syllabus5 domains · free

Amazon Web Services’s own domains. Each opens its own page listing the concepts beneath it, and each concept has a page of its own.

01Fundamentals of AI and MLThe vocabulary, the difference between artificial intelligence, machine learning and deep learning, the three ways a model learns, the stages of a machine learning pipeline, and how a model is judged once it exists.AI, machine learning and deep learning · Supervised, unsupervised and reinforcement learning · The machine learning lifecycle · Evaluating a machine learning model4 concepts
~20%
02Fundamentals of Generative AITokens, embeddings and transformers, what a foundation model can and cannot be relied on to do, and the lifecycle that takes one from raw data to a deployed system with feedback coming back.Foundation models and how they work · What generative AI is good and bad at · The foundation model lifecycle3 concepts
~24%
03Applications of Foundation ModelsThe heaviest domain. Choosing a model and setting its inference parameters, prompt engineering and the attacks against it, Retrieval Augmented Generation, the four ways to customise a model, and how to evaluate the result.Choosing and configuring a foundation model · Prompt engineering · Retrieval Augmented Generation · Customising a foundation model · Evaluating a foundation model5 concepts
~28%
04Guidelines for Responsible AIBias, fairness, robustness, safety and veracity as things you can measure and act on, the legal exposure that comes with generative output, and why an explainable model is sometimes worth a less accurate one.Responsible AI in practice · Transparency and explainability2 concepts
~14%
05Security, Compliance and Governance for AI SolutionsAccess control and encryption around a model, where training data came from and how you prove it, the standards an AI system is audited against, and the governance routine that keeps a deployed system accountable.Securing an AI system · AI governance and compliance2 concepts
~14%
Try a question

A logistics company has a system that decides whether a shipment needs a customs declaration. The logic was written by a compliance team as a set of conditions taken from published trade rules, and it is updated whenever the rules change. A stakeholder asks whether the company is already using machine learning. What is the accurate answer?

  • AYes. Any system that automates a decision a person used to make is machine learning by definition.
  • BYes, because the conditions were derived from real trade data held by the compliance team.
  • CNo. The behaviour was specified by people rather than learned from data, so it is artificial intelligence without machine learning.
  • DNo, and it is not artificial intelligence either, because no model is involved.
65 questions across the 5 domains.

One per page, with a worked explanation.

Start the set