TL;DR. Official AI certifications in 2026 are exams from Microsoft, Google Cloud, AWS, NVIDIA, Databricks, IBM, Salesforce and the IAPP. Anthropic keeps its own for partners. Microsoft retired AI-900, AI-102 and DP-100 in June 2026. Each vendor certification measures whether you know its platform; the AIGP measures governance. None of them measures how you use several AI tools day to day.
If you are working out which AI certification to sit this year, it is easy to end up preparing for an exam that no longer exists. On 12 November 2026, according to a Shakers check that day, Google's AI Overview for the search "best ai certifications" in the UK still placed the Microsoft Azure AI Engineer (AI-102) in its A-Tier, an exam Microsoft retired on 30 June 2026. The Overview for the broader "ai certifications" query already names AI-103 as its replacement. We call this a phantom certification. It shows up in the guides. You cannot sit it.
Last reviewed: 12 November 2026.
Which official AI certifications exist in 2026?
This map covers fifteen exam-based official AI certifications you can sit in November 2026, from nine issuers, counting Anthropic's programme as one. Microsoft contributes four (AI-901, AB-731, AI-103 and AI-300), AWS three and Google Cloud two. NVIDIA, Databricks, IBM and Salesforce each add their reference generative AI exam, and in governance there is the IAPP's AIGP. Anthropic awards four role-based credentials, but only to organisations in the Claude Partner Network. The table adds two credentials without a proctored exam: OpenAI's, limited to ChatGPT Enterprise and Edu, and Google's course certificate.
| Certification | Issuer | Level | Format | Status at 11/11/2026 | Source |
|---|---|---|---|---|---|
| Azure AI Fundamentals (exam AI-901) | Microsoft | Entry level | Proctored exam. Requires basic Python syntax | Current. Replaces AI-900 | Microsoft Learn |
| AI Transformation Leader (AB-731) | Microsoft | Entry level, business profile | 45-minute proctored exam, no coding | Current | Microsoft Learn |
| Azure AI Apps and Agents Developer Associate (AI-103) | Microsoft | Application and agent development | 120-minute proctored exam | Current. Replaces AI-102 | Microsoft Learn |
| Machine Learning Operations Engineer Associate (AI-300) | Microsoft | ML and MLOps | 120-minute proctored exam | Current. Replaces DP-100 | Microsoft Learn |
| Generative AI Leader | Google Cloud | Entry level, any role | 90 minutes, 50 to 60 multiple-choice questions | Current | Google Cloud |
| Professional Machine Learning Engineer | Google Cloud | ML | 2 hours, 50 to 60 questions | Current, with a new exam guide since 01/06/2026 | Exam guide |
| AI Practitioner | AWS | Entry level | 65 questions in 90 minutes | Current | AWS |
| Machine Learning Engineer Associate (MLA-C02) | AWS | ML | Beta with 85 questions in 170 minutes | In beta since 29/09/2026 | AWS blog |
| Generative AI Developer Professional | AWS | Advanced development | 75 questions in 180 minutes | Current | AWS |
| Generative AI LLMs Associate (NCA-GENL) | NVIDIA | Technical entry level | 1 hour, 50 to 60 questions | Current | NVIDIA |
| Generative AI Engineer Associate | Databricks | Development | 90 minutes, 45 scored questions | Current | Databricks |
| watsonx Generative AI Engineer Associate (C1000-185) | IBM | Development | 62 questions in 90 minutes | Current | IBM |
| Agentforce Specialist (AI-201) | Salesforce | Agents in CRM | 60 questions in 105 minutes | Current | Salesforce |
| AIGP (Artificial Intelligence Governance Professional) | IAPP | Governance | 100 questions in 2 hours and 45 minutes | Current | IAPP |
| Claude Certified (Associate, Developer and Architect) | Anthropic | By role | 120-minute proctored exam, closed book | Only for organisations in the Claude Partner Network | Anthropic |
| OpenAI Certified | OpenAI | Using ChatGPT | Courses and hands-on tasks inside ChatGPT, no proctored exam | Only ChatGPT Enterprise and Edu workspaces, by invitation | OpenAI Help Centre, checked on 11/11/2026 |
| Google AI Professional Certificate | Entry level, no prior knowledge | Programme of 8 courses on Coursera with assessments, no proctored exam | Available |
Here "official" means issued by the vendor or by the reference body for the discipline, not by a training academy.
Which AI certifications have been retired and what replaces them?
In 2026 Microsoft and AWS retired four AI exams that many guides still recommend: Microsoft's AI-900, AI-102 and DP-100, and the AWS Machine Learning Specialty; their replacements are AI-901, AI-103, AI-300 and the AWS Machine Learning Engineer Associate, in beta as MLA-C02. Microsoft retired DP-100 on 1 June and AI-900 and AI-102 on 30 June. AWS stopped offering the Machine Learning Specialty exam on 31 March. The TensorFlow Developer Certificate cannot be sat either: TensorFlow closed the exam.
| Retired exam | Last day | Replaced by | Source |
|---|---|---|---|
| AI-900, Azure AI Fundamentals | 30/06/2026 | AI-901, which awards the same certification | Microsoft Learn |
| AI-102, Azure AI Engineer Associate | 30/06/2026 | AI-103, Azure AI Apps and Agents Developer Associate | Microsoft Learn (same page) |
| DP-100, Azure Data Scientist Associate | 01/06/2026 | AI-300, Machine Learning Operations Engineer Associate | Microsoft blog |
| AWS Machine Learning Specialty (MLS-C01) | 31/03/2026 | AWS Machine Learning Engineer Associate | AWS |
| AWS MLA-C01, English version | 28/09/2026 | MLA-C02, in beta | AWS blog (see the map) |
| TensorFlow Developer Certificate | No published date | No replacement | TensorFlow |
The case that causes most confusion is AI-900, still the Microsoft AI certification exam many guides point to. The Azure AI Fundamentals certification still exists, but you now earn it with AI-901, which is more technical than AI-900: between 55% and 60% of the AI-901 exam is about implementing solutions with Microsoft Foundry. If you were after an entry exam with no code, look at AB-731 or Generative AI Leader instead.
Which AI certification should you take for your profile: entry level, development, ML or governance?
Which AI certification to take depends on your profile: without coding, Microsoft's AB-731, Google Cloud's Generative AI Leader or AWS AI Practitioner; with code, AI-901; for applications and agents, AI-103 or the AWS Generative AI Developer Professional; for ML, AI-300; and in governance, the AIGP. The cloud your projects run on decides the issuer. Outside the three clouds, NVIDIA's NCA-GENL covers foundations with code, while Databricks, IBM watsonx and Salesforce Agentforce cover applications and agents.
| Profile | Microsoft | Google Cloud | AWS | Other platforms |
|---|---|---|---|---|
| Entry level, no coding | AB-731 | Generative AI Leader | AI Practitioner | Google AI Professional Certificate (course) |
| Foundations with code | AI-901 | No equivalent in the map | No equivalent in the map | NVIDIA NCA-GENL |
| Applications and agents | AI-103 | No equivalent in the map | Generative AI Developer Professional | Databricks, IBM watsonx, Salesforce Agentforce |
| ML and MLOps | AI-300 | Professional Machine Learning Engineer | Machine Learning Engineer Associate | No equivalent in the map |
| Governance | No equivalent in the map | No equivalent in the map | No equivalent in the map | IAPP AIGP |
If you are starting out, do not aim for a Professional. Google recommends at least three years of industry experience, including one with Google Cloud, for its Professional Machine Learning Engineer. AWS asks for at least two years building production applications and one implementing generative AI for its Generative AI Developer Professional. That is putting the cart before the horse. If your goal is to build with AI and the foundations are still missing, get those in order first.
For agents specifically, GitHub also has its own credential, focused on agentic development.
In governance, the IAPP's AIGP is the reference credential for individuals: an exam on the development, deployment and responsible management of AI systems. The ISO/IEC 42001 standard, for AI management systems, is something else. ISO itself states that it does not certify or issue certificates, so Lead Implementer and Lead Auditor titles are awarded by private bodies, not by ISO.
What does each certification measure, and what does none of them measure?
Vendor AI certifications measure whether you know a platform: its services and how a solution is deployed on it. The AIGP measures whether you know AI governance. They check it with a proctored exam, a course with assessments or, at OpenAI, tasks inside ChatGPT. None of them measures how you use AI day to day with several tools at once. Not which agents you have running, not how you split work between one coding agent and another, not what you check before signing something off.
| Type of signal | How it assesses | What it shows | What it leaves out |
|---|---|---|---|
| Vendor exam (Microsoft, Google Cloud, AWS, NVIDIA, Databricks, IBM, Salesforce) | Proctored exam with multiple-choice questions and scenarios | Knowledge of a platform's services and architecture | Daily use and every other tool |
| Tool credential (OpenAI, Anthropic) | Tasks inside ChatGPT or a role-based exam on Claude | Command of one tool or one model | What you do outside that tool |
| Course certificate (Google) | The course's own assessments | That you completed a training programme | Sustained use outside the course |
| Governance (IAPP AIGP) | Proctored exam | Knowledge of governance and responsible use of AI | Technical practice |
| Usage credentials on LinkedIn (Descript, Lovable, Relay.app, Replit) | Real usage patterns inside each tool | Skill with that tool | The rest of the stack |
| AI fluency level (Shakers) | Usage data from the Shakers MCP or CLI, or a technical interview with Alma | How you use AI every day across several tools | Knowledge of one specific platform |
That day-to-day work no longer fits inside a single tool. According to the State of European AI Builders 2026 study by Shakers (1,035 tech professionals, 23 European countries), 73% keep at least one AI agent or automated workflow running in real work, and weekly use is spread across ChatGPT (70%), Claude (57%), Gemini (51%) and Claude Code (43%), in a multiple-answer question. You can download the full study.
The two most hands-on credentials come with small print. OpenAI's stays inside ChatGPT and only reaches Enterprise and Edu workspaces. Anthropic's, according to the FAQ for its partner programme, is only for the Claude Partner Network. An independent tech professional cannot sit either of them.
The closest thing to measuring real usage came from elsewhere. Since 26 January 2026, LinkedIn has let profiles display credentials from Descript, Lovable, Relay.app and Replit, which each company issues based on real usage inside its product, according to LinkedIn's announcement. But each one looks at a single tool. And an AI Builder rarely lives inside just one.
Are there free AI certifications?
Not among exam-based AI certifications: no exam in the map is free. What is free is the preparation and, in some cases, the upkeep. Microsoft offers the training for its exams on Microsoft Learn and renews its certifications at no cost. AWS includes free foundational AI training in its exam prep plans. And OpenAI Academy has free courses, although OpenAI makes clear that its badges are not certifications. An AI certification course that ends in a completion certificate is not the same as a proctored exam. Many lists mix the two.
| Certification | Exam price published by the issuer | Validity |
|---|---|---|
| Azure AI Fundamentals (AI-901) | Depends on the exam country | Does not expire |
| AI Transformation Leader (AB-731) | Depends on the exam country | Not stated |
| AI-103 and AI-300 | Depends on the exam country | One year, with free online renewal |
| Generative AI Leader | 99 US dollars | 3 years |
| Professional Machine Learning Engineer | 200 US dollars | 2 years |
| AWS AI Practitioner | 100 US dollars | 3 years |
| AWS Machine Learning Engineer Associate | 75 US dollars for the MLA-C02 beta | 3 years |
| AWS Generative AI Developer Professional | 300 US dollars | Not stated |
| NVIDIA NCA-GENL | 125 US dollars | 2 years |
| Databricks Generative AI Engineer Associate | 200 US dollars per attempt | 2 years |
| IBM watsonx Generative AI Engineer Associate | 200 US dollars | Not stated |
| Salesforce Agentforce Specialist | 200 US dollars, and 100 to retake | Not stated |
| IAPP AIGP | 649 US dollars for members and 799 for non-members | 2 years, with 20 continuing education credits |
| Claude Certified | From 99 to 175 US dollars depending on the credential | 12 months |
| OpenAI Certified | Not published | Not stated |
Amounts are in US dollars, as published by each issuer on the pages linked in the map, in the IAPP store and in the Google Cloud help centre, and several note that tax is added on top. Microsoft sets its price per country. The Google AI Professional Certificate is not listed because it is a course, not an exam, and it is not free either: the programme page on Coursera lists 49 US dollars a month for the US and Canada.
How does a vendor certification complement a real usage level?
They complement each other because they answer different questions. Anyone assessing an AI Builder needs both:
- The vendor certification proves, through an exam, what you know about a platform such as Azure, Google Cloud or AWS.
- The real usage level shows which tools and agents you use when you collaborate on projects, not just on exam day.
At Shakers we call it the AI fluency level: it is calculated from usage data from the Shakers MCP or CLI, or from a technical interview with Alma, our AI agent, inside the talent infrastructure Shakers uses to certify skills. It is worth saying plainly, because it is ours: it does not replace any exam and does not claim to be one.
That is what moving from roles to skills and agents means: a job title says little if nobody can see what you do with AI. If you already hold a certification, or are preparing for one, measure how you use AI as well.
Frequently asked questions about AI certifications
- What AI certifications are there?
In November 2026 you can sit official AI certifications from Microsoft (AI-901, AB-731, AI-103 and AI-300), Google Cloud (Generative AI Leader and Professional Machine Learning Engineer), AWS (AI Practitioner, Machine Learning Engineer Associate and Generative AI Developer Professional), NVIDIA, Databricks, IBM, Salesforce and the IAPP's AIGP. Anthropic and OpenAI have their own credentials, but they are not open to everyone.
- What is the best AI qualification?
There is no single best one for everyone: it depends on your profile and on the cloud your projects run on. Without coding, AB-731, Generative AI Leader or AWS AI Practitioner; to build applications and agents, AI-103 or the AWS Generative AI Developer Professional; for ML, AI-300; for governance, the IAPP AIGP. Whichever you pick, check that it ends in a current exam, not a retired one such as AI-900 or AI-102.
- Is a certificate in AI worth it?
It is worth it when it matches the platform your projects run on and the exam is still live. A vendor certification proves, through a proctored exam, what you know about a platform such as Azure, Google Cloud or AWS. It does not show how you use AI day to day across several tools, so anyone assessing your profile will also want to see that.
- How can I get certified in AI?
Choose the issuer by the cloud or platform your projects use, prepare the syllabus with the official training and book the proctored exam, online or at a test centre: AI-901, AWS AI Practitioner and the AIGP, for example, are booked through Pearson VUE. Before paying, read the official page: duration and recommended experience vary a lot. If you are starting without coding, Microsoft's AB-731 and Google Cloud's Generative AI Leader ask for less technical grounding.
- What free AI courses with a certificate are available?
The free courses from the issuers in the map give you training, not certification. Microsoft Learn offers exam preparation at no cost, AWS includes free foundational AI training in its exam prep plans and OpenAI Academy has free courses whose badges, according to OpenAI, are not certifications. None of them replaces the exam, which is paid for every exam-based certification in the map. And the Google AI Professional Certificate is not free: it is a paid programme of courses on Coursera.
- Is AI-900 still valid?
You can no longer sit it: Microsoft retired it on 30 June 2026. Anyone who passed it keeps the Azure AI Fundamentals certification, because retiring an exam does not invalidate certifications already earned and Microsoft's Fundamentals do not expire. If you want that same certification today, the exam is AI-901, which requires basic Python and devotes more than half the exam to implementing solutions with Microsoft Foundry.
- What is the difference between AI-900 and AI-901?
They award the same certification, Azure AI Fundamentals, through different exams. AI-900 was retired on 30 June 2026 and AI-901 is the current one. AI-901 measures two areas: AI concepts and capabilities, between 40% and 45% of the exam, and implementing solutions with Microsoft Foundry, between 55% and 60%. It also requires basic Python syntax.
- What is the difference between AI-102 and AI-103?
AI-102 awarded the Azure AI Engineer Associate certification and Microsoft retired it on 30 June 2026, according to its table of retired exams on Microsoft Learn. Its replacement, AI-103, awards a different certification, Azure AI Apps and Agents Developer Associate, aimed at people who build, manage and deploy agents and AI solutions with Microsoft Foundry. The exam is proctored and lasts 120 minutes, and the certification is renewed every year for free.
- Do AI certifications expire?
It depends on the issuer. Microsoft's Fundamentals do not expire and its Associate certifications are renewed every year with a free online assessment. AWS AI Practitioner and Machine Learning Engineer Associate last three years. At Google Cloud, Professional Machine Learning Engineer lasts two and Generative AI Leader three. NVIDIA's NCA-GENL and the Databricks certification last two years, the AIGP is maintained for two with 20 continuing education credits and Anthropic's last 12 months.
- Is there a certification specifically for AI agents?
Several in the map cover agents, such as Microsoft's AI-103 and Salesforce's Agentforce Specialist. GitHub also has a credential focused on agentic development.
