AWS Machine Learning Specialty Is Retired: What Replaces It in 2026

AWS spent more than seven years building the Machine Learning Specialty certification into the credential serious ML practitioners chased. Then, this year, AWS quietly closed it. The last exam sat on March 31, 2026, and there is no direct successor. Instead, AWS split that territory across four separate certifications, each aimed at a different kind of AI work.

If you had Machine Learning Specialty on your study plan for this year, that plan needs a rewrite. This guide walks through why AWS made the change, what each of the four replacement paths actually tests, what they cost, and which one fits the job you actually do, backed by real exam specs and current labor market data rather than a generic “which cert should I get” checklist.

Why AWS Pulled the Plug on Machine Learning Specialty

When Machine Learning Specialty launched, most organizations sitting for it were still building and tuning models from scratch. Data scientists trained algorithms, tuned hyperparameters, and evaluated models largely by hand. That was the discipline the exam was built to certify.

That is not how most AI work happens now. Teams increasingly start from a managed foundation model through Amazon Bedrock or SageMaker’s built-in model catalog and spend their engineering time on integration, deployment, and governance rather than training a model from a blank slate. A single specialty-level exam could not cleanly cover both the old discipline and the new one, so AWS split it in two: one path for people who still do hands-on ML engineering, and one for people who build applications on top of foundation models.

A few things are worth knowing, confirmed directly on AWS’s certification status page for the retired exam:

• The last day to take AWS Certified Machine Learning Specialty was March 31, 2026. It is not being offered anywhere at any testing provider after that date.
• If you already passed it, nothing changes. Your certification stays active for three years from the date you earned it, exactly as it always would have.
• Your digital badge keeps working. It stays visible and shareable through Credly for the full three-year window.
• AWS is not issuing a single 1:1 replacement. Official guidance points holders toward four different certifications depending on what they actually do day to day.

The Four AWS AI Certifications Filling the Gap

AWS’s training and certification team laid out the transition plan in a portfolio update this year, naming AI Practitioner, Machine Learning Engineer Associate, Data Engineer Associate, and the new Generative AI Developer Professional as the landing spots for people who would previously have sat Machine Learning Specialty. None of them is a straight swap. Here is how they actually stack up against each other.

AWS Certified AI Practitioner (AIF-C01): Start Here With Zero Prerequisites

AI Practitioner is the foundational credential in the new lineup, and it is deliberately built for people who work alongside AI systems without necessarily building them: product managers, analysts, sales engineers, and career-switchers moving into AI-adjacent roles. AWS states plainly that the ideal candidate is “familiar with, but does not necessarily build” AI and ML solutions on AWS, which is a very different bar than Machine Learning Specialty ever set.

The exam breaks down into five weighted domains: Fundamentals of AI and ML (20%), Fundamentals of Generative AI (24%), Applications of Foundation Models (28%), Guidelines for Responsible AI (14%), and Security, Compliance, and Governance for AI Solutions (14%). Notice that the single heaviest domain is applying foundation models rather than building them from scratch, which lines up exactly with the shift described above. If you want to see the exact question style before booking a seat, PracticeTestSoftware’s AI Practitioner practice test mirrors that domain weighting rather than treating every topic as equally likely to appear.

One detail that surprises people: AI Practitioner is valid for three years, and you can recertify either by retaking the current version of the exam or by earning Machine Learning Engineer Associate instead. That is AWS explicitly building a bridge from the foundational cert into the more technical one, which tells you something about how it expects most holders to progress.

AWS Certified Machine Learning Engineer, Associate: The Hands-On Successor

If your actual job involves training, tuning, deploying, and monitoring models, this is the closest thing to a direct heir to Machine Learning Specialty, just recalibrated to associate-level scope. AWS recommends at least a year of hands-on experience with Amazon SageMaker and related ML engineering services before attempting it. The exam runs 130 minutes across 65 questions and costs $150, noticeably more expensive and more technical than AI Practitioner.

What Changes When MLA-C01 Becomes MLA-C02

This certification is itself mid-transition right now, on a timeline that matters if you are planning to sit it this year:

• September 1, 2026: registration opens for the MLA-C02 beta exam, English only.
• September 28, 2026: the current English-language MLA-C01 stops being offered.
• At general availability: the updated MLA-C02 exam adds Korean, Japanese, and Simplified Chinese alongside English.

Anyone planning to sit this exam soon should book before late September or plan around the beta window instead, since the two versions will not run side by side indefinitely.

AWS Certified Generative AI Developer, Professional (AIP-C01): Built for the Bedrock Era

This is the newest credential in AWS’s AI lineup and the most demanding one. It moved from beta into standard registration this year, and the standard version was refreshed to include Amazon Bedrock AgentCore, AWS’s newer framework for building and running autonomous agents in production. The exam covers integrating foundation models into real applications and workflows, including retrieval-augmented generation architectures and vector database design, not just calling an API and printing the result.

At $300 for 75 questions across 180 minutes, it is priced and timed like the professional-level exam it is. AWS recommends two or more years building production applications on AWS plus at least a year of hands-on generative AI implementation work, and notes candidates tend to do better if they already hold AI Practitioner, Solutions Architect, Machine Learning Engineer Associate, or Data Engineer Associate first. Alongside the exam, AWS also introduced a companion Agentic AI Demonstrated microcredential, a hands-on, lab-based way to prove you can actually build an agent rather than just answer questions about one. If you already have a Bedrock project shipped, PracticeTestSoftware’s Generative AI Developer Professional practice questions are a reasonable way to check whether your knowledge gaps are conceptual or just exam-format related.

AWS Certified Data Engineer, Associate (DEA-C01): The Path for Pipeline Builders

Data Engineer Associate is the option AWS points to for anyone whose actual work is closer to building the pipelines that feed models than building the models themselves. It tests data ingestion, storage, transformation, security, and monitoring across services like AWS Glue, Amazon Redshift, EMR, Kinesis, and S3, plus the IAM, KMS, and Lake Formation layers that keep that data governed properly. AWS recommends 2 to 3 years of data engineering experience and 1 to 2 years of hands-on AWS work before attempting it.

It runs $150 for 65 questions in 130 minutes, the same format as Machine Learning Engineer Associate, which makes sense since both sit at associate level. For a fuller walk-through of the domains and services this exam actually tests, PracticeTestSoftware’s own Data Engineer Associate exam roadmap breaks the ingestion and transformation domains down service by service.

What Companies Are Actually Paying for AI Skills in 2026

The retirement of one exam matters less than the demand sitting underneath all four replacements, and that demand is real, not marketing copy. The U.S. Bureau of Labor Statistics projects data scientist employment will grow 34% from 2024 to 2034, expanding from about 245,900 jobs to 328,300, which makes it the fourth-fastest-growing occupation the agency tracks and the fastest-growing one in its mathematical science category. BLS also expects roughly 23,400 openings per year over that decade, and puts the 2024 median wage for the role at $112,590.

That is broader labor market context, not a certification-specific number, and it is worth being honest about that distinction. But it explains why AWS is investing in four certifications instead of maintaining one.

On the employer side, a 2026 workforce study AWS ran with the policy research firm Access Partnership, surveying 3,297 employees and 1,340 organizations across the U.S., found employers willing to pay real premiums for verified AI skills, varying by function:

• IT roles: up to 47% more
• Sales and marketing: 43% more
• Finance: 42% more
• Business operations: 41% more
• Legal, regulatory, and compliance: 37% more
• Human resources: 35% more

The same study found close to 80% of workers surveyed said they wanted to build AI skills for exactly this reason. None of that guarantees any individual a raise for passing an exam, but it does mean the underlying skills these four certifications test are ones employers are actively budgeting for right now, across departments that have nothing to do with engineering.

The Security Specialty Also Got an AI Makeover

Machine Learning Specialty was not the only exam AWS touched this year. AWS Certified Security, Specialty moved to a new version, SCS-C03, after the prior version’s testing window closed on December 1, 2025. It is not a Machine Learning Specialty replacement and AWS has never framed it as one, but it is part of the same portfolio shake-up and worth knowing about if AI security work is anywhere near your role.

The updated exam restructures its domains to put more weight on detecting and responding to security incidents in AI and ML workloads specifically, rather than treating AI systems as just another workload type to secure the same way as everything else. It still targets experienced security professionals, AWS recommends five years of general IT security experience and at least two years securing AWS workloads specifically, and it remains a 170-minute, 65-question exam priced at $300. If cloud security is your world, it is a useful signpost that AI-specific threats are now formally part of what a “security specialist” is expected to know, not a side topic.

AWS is not alone in reshaping a cloud security credential around AI this year. ISC2 pushed its own 2026 update to the Certified Cloud Security Professional exam outline for largely the same reason, so this is turning into an industry-wide pattern rather than something specific to one vendor.

So Which One Should You Actually Sit For?

Strip away the marketing and the decision mostly comes down to what you actually do at work today, not what sounds impressive on a resume.

• Pick AI Practitioner if you work with AI-touched products or teams but do not build models yourself, or if this is your first AWS AI certification and you want a low-cost, no-prerequisite entry point.
• Pick Machine Learning Engineer Associate if you are hands-on with SageMaker, training pipelines, or model monitoring, and Machine Learning Specialty was the exam you were actually planning to take.
• Pick Generative AI Developer Professional if you are shipping applications built on Bedrock or building agentic systems, and you already have real production experience to draw on, not just tutorial-level exposure.
• Pick Data Engineer Associate if your actual job is building and securing the pipelines that feed data into models, rather than the modeling work itself.

It is also fair to hold more than one. AWS explicitly designed AI Practitioner to feed into Machine Learning Engineer Associate for recertification purposes, and the professional-level Generative AI Developer exam explicitly rewards candidates who already hold one of the associate-level certifications first.

Where PracticeTestSoftware Fits Into This

Worth being upfront about: PracticeTestSoftware is an independent certification prep provider, not Amazon Web Services, and has no affiliation with, endorsement from, or sponsorship by AWS. Exam pricing, retirement dates, and blueprint details in AWS’s certification program move fast, so treat the specifics above as accurate as of this writing and double-check anything time-sensitive on AWS’s own certification pages before you register or pay for an exam.

Frequently Asked Questions About AWS’s 2026 AI Certification Overhaul