AWS Certified Machine Learning – Specialty (MLS-C01) is no longer an exam that a new candidate can schedule. AWS ended testing on March 31, 2026. The knowledge associated with the certification remains useful, but the distinction between a retired assessment and the skills it once validated matters: an old study plan is not a current booking route, and a historical exam code should not be linked to a commercial purchase as though registration were still available.
The former specialty certification assessed experienced practitioners who designed, trained, tuned, deployed and maintained machine-learning workloads on AWS. It required more than recognition of AWS service names; it connected data suitability, modeling choices, operational concerns and cloud architecture. Today, that professional knowledge can inform modern ML engineering, but candidates choosing a new AWS credential need to assess the available programs on their own terms rather than assume a direct, identical successor exists.
What MLS-C01 represented in AWS Certification
MLS-C01 was a specialty-level examination, rather than a prerequisite for an entry-level cloud role. AWS targeted people with substantial practical experience developing and operating ML solutions. Its historical public exam description listed 65 multiple-choice or multiple-response questions and a 180-minute testing period. Those are archival facts about MLS-C01, not instructions for an active appointment.
A professional who earned the credential before retirement does not automatically lose it on March 31, 2026. AWS says the certification remains active for three years from the date earned, including previously issued digital badges. Retirement stops new testing and recertification through that retired assessment; it does not retroactively invalidate an already-earned qualification before its normal expiration.
For readers comparing current paths, the broader AWS AI and machine-learning landscape now includes role-based learning and new work in generative AI. The right next step may be an engineering credential, deeper system implementation or practical proof of competence, depending on what the person wants to build.
Read the historical four-domain blueprint in context
The former MLS-C01 blueprint allocated 20% to Data Engineering, 24% to Exploratory Data Analysis, 36% to Modeling, and 20% to Machine Learning Implementation and Operations. These percentages explain the legacy specialty’s priorities. They must not be relabeled as the 2026 Machine Learning Engineer – Associate content outline, which has different domains and an active version transition.
Data Engineering concerned the ability to acquire, transform and move data into a usable state. A suitable design might land records in Amazon S3, process them with an appropriate transformation service, protect sensitive fields, and deliver a reproducible dataset to a training workflow. The service choice depended on ingestion patterns, latency, format, volume and governance—not whether the candidate happened to know one service’s marketing description.
Exploratory Data Analysis focused on understanding the data before fitting a model. A useful practitioner checks sampling, missingness, label quality, imbalance, leakage and the relationship between features and outcomes. Treating a high training score as proof of a valid model can conceal target leakage or an inappropriate validation split. That concern remains relevant in both traditional ML and modern AI applications.
Modeling covered matching algorithms and evaluation to the problem. Forecasting, fraud classification and recommendation tasks have different error costs and selection criteria. The largest historic domain was not an invitation to memorize every mathematical formula; it examined how technique, data, evaluation and deployment constraints affected the chosen solution.
Implementation and Operations considered what happens after modeling. Versioning, deployment patterns, monitoring, data drift, retraining triggers, scaling, security and cost control turn an experiment into a service. The ML pipeline lifecycle provides a useful bridge between historical exam domains and the ongoing responsibilities of an ML engineer.
One practical case illustrates why the skills still matter
Consider a manufacturer trying to predict equipment failure early enough to plan maintenance. Sensors produce time-series events, work orders provide imperfect labels, and the maintenance team needs a model that distinguishes costly failures from harmless anomalies. The technical problem is not solved by selecting ‘machine learning’ in a product console. The team must understand data collection, timeliness, false positives, operational response and the real cost of missed events.
An initial data engineering stage must preserve event time, asset identity and maintenance history. If the pipeline mixes records from after a failure into a training row intended to predict the failure beforehand, the measured model performance is not trustworthy. A defensible evaluation uses splits and backtesting that respect how the model would be used in production.
Model selection then depends on constraints. A simple baseline might be preferable to a complex neural network if it is robust, explainable and easier to maintain. A classifier with excellent aggregate accuracy can still be unsafe if rare but important failures are missed. Decision thresholds should therefore incorporate the maintenance team’s ability to respond and the consequences of false alarms.
After deployment, the team tracks input quality, latency, prediction patterns, downstream outcomes and changes in failure rates. If the model no longer performs, the cause may be drift, an upstream schema change or an operational workflow that stopped acting on alerts. Those interdependent controls express the real competence that historic MLS-C01 preparation was trying to measure.
Understand the MLA-C01 to MLA-C02 transition accurately
AWS Certified Machine Learning Engineer – Associate is a separate, associate-level credential. As of October 27, 2026, its updated MLA-C02 examination is in English beta, not yet generally available. Beta delivery began September 29, 2026, after registration opened September 1. AWS announced general availability for January 14, 2027. It would be wrong to label the beta as already in general release during October.
English MLA-C01 testing ended September 28, 2026, but Korean, Japanese and Simplified Chinese delivery of MLA-C01 continues through the transition until general availability of MLA-C02. This language-specific detail is important: a universal ‘MLA-C01 retired in every language’ statement would be false in October 2026.
The updated associate exam includes traditional ML engineering as well as foundation models, generative AI, Amazon Bedrock, retrieval-augmented generation and agentic workflows. An MLA-C02 preparation approach should follow its official beta exam guide rather than inherit the specialty’s old four-domain percentages.
These credentials should not be treated as equivalent levels or automatic replacements. MLS-C01 historically validated specialty expertise across the full ML workflow; the associate pathway focuses on the implementation and operation of ML and AI solutions for its defined target role. Compare the actual examination scope with your experience and goals before choosing what to study.
How current AWS services fit the preserved knowledge
Amazon SageMaker AI and related AWS tooling can still help teams prepare datasets, train models, evaluate outcomes and manage production endpoints. The specific tool should be selected according to workload requirements: managed features may reduce undifferentiated operational work, while custom containers and workflows may be appropriate for requirements that cannot be met by defaults.
An introduction to Amazon SageMaker provides one conceptual starting point, but an experienced practitioner also needs to understand IAM policies, encryption, private connectivity, data retention and observability. A model that delivers excellent predictions but exposes sensitive training data fails an essential production obligation.
Modern generative AI systems introduce further constraints. Selecting a foundation model is not the same as proving that its output is grounded, safe and useful. Retrieval systems need document access control and relevance evaluation; agents need narrowly defined tool authority and audit trails. The historic specialty did not test the present MLA-C02 generative-AI scope, but its insistence on data quality, deployment discipline and monitoring remains relevant.
What existing holders and new learners should do
An existing MLS-C01 holder should retain the original award date, preserve evidence of active status through the credential’s normal expiration and describe the certification accurately in a CV or professional profile. Stating that the exam retired while a particular person’s credential remains active is a meaningful distinction. Employers should evaluate current practice alongside the badge, because tools and governance requirements continue to evolve.
A learner who did not earn MLS-C01 should stop pursuing old exam registration instructions, discounted retired-exam listings or outdated practice questions. Build a portfolio that demonstrates ingestion, leakage-resistant validation, appropriate model evaluation, reproducible deployment and post-release monitoring. Then compare the present associate or other active AWS role-based credentials with the work you want to perform.
The AWS certification progression landscape can help frame those decisions, but a path should be chosen from current issuer status and job responsibilities rather than the superficial similarity of exam names. Retirement changed the credential options; it did not eliminate the professional value of sound machine-learning engineering.
A practical decision starts with the responsibilities the learner wants to own. Someone building production pipelines should demonstrate reproducible transformations, controlled deployment and monitoring. Someone evaluating AI products needs a different balance of technical depth, business risk and responsible-use judgment. The best current certification is the one whose official task list matches those duties, not simply the badge with the most advanced-sounding name. Check that the exam is actually available in your preferred language and delivery period. Certification status, examination version, course availability and a specific person’s still-valid award are separate facts that should be verified independently before money or study time is committed.
What this historical certification still teaches
MLS-C01 provides a useful archive of how AWS once assessed end-to-end cloud ML judgment: trustworthy data, defensible modeling, deployable systems and operational evidence. Its March 2026 retirement is final for new testing. The appropriate modern response is to preserve the historical record, validate any still-active individual credential and develop current skills through the live exam guide and practical ML systems.