Purpose & scope
CAIP certifies applied judgement and diagnostic competence in building and operating LLM- and agent-based systems: choosing between prompting, retrieval, model substitution and fine-tuning; designing agent loops and tool integrations; building evaluations; diagnosing failures from traces; controlling cost and latency; and constraining what an autonomous system is permitted to do.
The scope is applied LLM applications and agents. It does not cover classical machine learning, computer vision, statistical modelling or model-training research.
What it does not certify
CAIP does not attest to hands-on build proficiency. The exam assesses judgement and diagnosis, not construction. An employer should treat a pass as evidence that the holder can reason correctly about production AI systems and diagnose them from evidence — not as proof that they have built one to a given standard.
Who it is for
Practitioners who build with models, retrieval and agents: AI and software engineers, ML and platform engineers, technical leads and solutions architects. Entry-level programming ability is recommended. Any individual aged 18 or over may apply; you need an account, a verified email, and identity verification for the proctored exam.
Competence domains (exam blueprint)
The exam is weighted across seven domains: Models, Prompting and Context (16%), Retrieval and Data Lifecycle (11%), Agents and Reliable Integrations (21%), Evaluation and Observability (15%), Security, Privacy and Authority (15%), Software Engineering and Operation (12%), and Problem Framing, Economics and Delivery (10%).
Each question maps to a published learning objective. Most questions supply an artefact — an execution trace, retrieved passages, a cost table, a configuration or a log excerpt — that you must read to answer.
Exam format
60 four-option multiple-choice questions, 130 minutes, delivered online with identity verification and proctoring. Each candidate receives a freshly assembled, individually randomised form. Every question is reviewed by a subject-matter expert before it can appear on a form.
The pass mark is criterion-referenced — set by a standard-setting study against the published competence standard, not by ranking candidates against each other — and is shown on the exam page before you begin. The exam is not administered until that standard has been set.
The study uses the modified Angoff method: a panel of qualified practitioners, each declaring any conflict of interest, estimates for every question how many of 100 just-competent candidates would answer it correctly; the averages, rounded to the nearest whole percent, become the pass mark, which the Certification Committee approves before use. Your attempt is marked against the pass mark in force when you begin, and a later change never affects an attempt already started. The full explanation, with a worked example, is published at /certifications/certified-ai-practitioner/pass-mark.
Vendor neutrality
Questions may name a protocol or a category of tool where that is the accepted term of art, and never depend on a particular vendor's interface, version, pricing or defaults. A candidate who has only worked on a competing stack can score full marks.
Preparation
AIPA publishes the preparation course free of charge. It is optional: your result depends only on your exam performance against the published standard, never on whether you studied with us.
Result & credential
On a pass, AIPA issues a verifiable digital credential with a unique ID that anyone can confirm on the public verification page or via the verification API. The credential states its exact name, level and validity dates.
Validity & recertification
The credential is valid for one year and is renewed annually — by continuing education or by passing the exam again — as set out in the Recertification Policy.
Earlier edition
Version 1.0 of this scheme described a Foundation-level CAIP (five domains, 80 questions, 120 minutes). That edition was discontinued on 2 November 2026; no credentials issued under it remain in use.
Fees
CAIP lists at US$299. A time-limited founding-cohort price of US$99 applies to the first 500 candidates certified. Fees are published and applied equally to all candidates.
Rights & responsibilities
Candidates have the right to a fair, impartial assessment, reasonable accommodations, and access to appeals and complaints. Candidates are responsible for exam integrity and truthful use of the credential (see the Candidate Agreement and Mark Use Policy).
This policy is provided for transparency and is not legal advice. Where local law grants you stronger rights, those rights prevail.