Foundation certification

UKAIC AI Foundation

The entry-level certification in the UKAIC professional AI certification pathway. A vendor-neutral standard of AI literacy and responsible professional AI use, for candidates in any industry, at a level that requires no coding, mathematics, or technical background.

£150

Includes one examination attempt

What you’re buying

This purchase is for the examination only. Training that prepares candidates for it is delivered separately, by authorised training providers — completing any training activity does not itself confer this certification. It is awarded solely on passing the exam.

UKAIC AI Foundation is an industry-recognised certification issued by an independent membership organisation. It is not a regulated qualification and does not sit on a national qualifications framework.

Who it’s for

The UKAIC AI Foundation is the entry-level certification in the UKAIC professional AI certification pathway. It establishes a vendor-neutral standard of AI literacy and responsible professional AI use for candidates in any industry.

The certification tests a candidate's understanding of what AI is and how it produces outputs; their ability to use AI effectively and appropriately; their capacity to critically evaluate and verify AI output; their protection of information; and their responsible, accountable use of AI — all at a level that requires no coding, mathematics, or technical background.

Exam format

Questions: 120 (Multiple choice, single best answer)
Pass mark: 70% (84 of 120)
Duration: 120 minutes
Delivery: Supervised / proctored, administered by authorised providers
Book: Closed book
Prerequisites: None
Vendor neutrality: No question requires knowledge of any specific commercial AI product

What’s tested

Domain 1

Understanding AI: The Big Picture

12% · 14Q

Establishes accurate, vendor-neutral vocabulary and framing for artificial intelligence.

1.1

Defining AI

What artificial intelligence is: most modern AI learns patterns from data to make predictions, generate content, or take actions.
What AI is not: not conscious, not infallible, not "magic".
That some systems are rule-based rather than learning systems, and that a simple explanation can still be accurate ("simple ≠ wrong").

1.2

AI, ML, Deep Learning & Generative AI

How the terms nest: AI contains machine learning, which contains deep learning, which contains generative AI and large language models.
Distinguishing the four terms and recognising when they are wrongly used as synonyms.
Why precision matters when reading tool claims, vendor marketing, and media coverage.

1.3

Narrow vs General AI

Narrow AI: systems designed for specific tasks, capable within their domain but limited outside it.
General AI (AGI): a hypothetical system able to perform any intellectual task a human can — a research goal, not a current product.
That all commercial AI today is narrow, and applying the "narrow or general?" test to dramatic claims.

1.4

Everyday AI

Recognising narrow AI already embedded in routine tools: spam filtering, search ranking, navigation, and recommendations.
Understanding that most professionals are already daily AI users.

Domain 2

How AI Works (Without the Maths)

14% · 17Q

An accurate, non-mathematical model of how AI produces outputs and where it is unreliable.

2.1

Models, Training & Inference

A model as a set of learned values (parameters), not a database of stored answers; the role of datasets.
The distinction between training (learning the model) and inference (using it to produce an output).
Parameters at Foundation level: the adjustable values learned during training.

2.2

How Language Models Generate Text

Next-token prediction: producing likely continuations from the prompt and text so far.
That output is generated, not retrieved, and that fluency does not guarantee accuracy.
That the mechanism is more sophisticated than simple autocomplete, yet has no built-in fact-checker.

2.3

Tokens & the Context Window

Tokens as the basic unit of text a model reads and writes.
The context window as short-term working memory, and the effects of exceeding it (earlier detail "forgotten", large pastes crowding out instructions).
Practical ways to work within the window (concise prompts, key instructions first/last, chunking).

2.4

Knowledge Cut-off

That a base model has no knowledge of events after its training date.
Why time-sensitive questions are unreliable from the base model alone.
That some applications add live search, but the base model itself remains frozen.

2.5

Variability & Embeddings

Why identical prompts can produce different answers (sampling / the "temperature" setting).
The distinction between a model and the application built around it.
Embeddings as the representation of meaning as numbers, so related items sit close together. (awareness)

Domain 3

The AI Landscape & Generative AI

12% · 14Q

A vendor-neutral map of AI capabilities and how to choose the right category of tool.

3.1

Generative AI & Output Types

The meaning of "generative" (producing new content) versus classifying or labelling.
The major output types: text, image, audio/voice, video, and code.

3.2

Foundation Models & Multimodal AI

Awareness level

Foundation models as large, general-purpose models adapted to many tasks. (awareness)
Multimodal AI as handling more than one type of input or output. (awareness)

3.3

Retrieval-Augmented Generation (RAG)

The grounding problem, and RAG as retrieving relevant sources then generating an answer grounded in them.
That RAG improves freshness and grounding but reduces rather than removes error.

3.4

AI Agents

Agents as multi-step, tool-using systems that pursue a goal, distinct from a single-turn assistant.
Why agents that take actions require greater human oversight. (awareness)

3.5

Model, Application & Service

Distinguishing the model (engine), the application (interface), and the AI-enabled service.
Open-weight / self-hosted versus hosted options and their trade-offs (control and data residency versus setup and maintenance). (awareness)

3.6

Selecting a Tool Category

Choosing an appropriate category of tool — not a specific product — using capability, cost, data sensitivity, and organisational approval.

Domain 4

Prompt Engineering Fundamentals

16% · 19Q

The highest-leverage practical skill: constructing effective prompts and improving weak ones.

4.1

Anatomy of a Prompt

The four components — task, context, format, and constraints — and how to apply them to construct an effective prompt.

4.2

Role & Audience

Assigning a role or persona to the model and naming the audience to shape the register and depth of output.

4.3

Few-shot Prompting

The difference between zero-shot and few-shot prompting.
Using one or more examples to calibrate style, format, and tone, and when to do so.

4.4

Structure & Decomposition

Requesting structured output (e.g. tables, headed sections).
Decomposing a complex task into ordered steps, and asking for a plan or reasoning where useful.

4.5

Iterative Refinement

Treating the first output as a draft and refining through specific, corrective follow-ups.
Understanding prompting as a conversation that builds on previous responses.

4.6

Common Prompting Faults

Diagnosing and correcting vague, overloaded, leading, unformatted, and "one-and-done" prompts.

Domain 5

Working with AI Day to Day

12% · 15Q

Applying AI to real professional work, with the judgement and habits that keep it safe.

5.1

Drafting & Rewriting

Using AI to produce first drafts and to reshape content (shortening, re-levelling for an audience, changing tone or format).
Reviewing and personalising AI output rather than using it unchanged.

5.2

Summarising

Producing directed and audience-targeted summaries.
The limitation that summaries can omit critical detail, and the need to read the original for high-stakes documents.

5.3

Ideation, Planning & Research Assistance

Using AI as a thinking partner for ideas and planning, with human ownership of the result.
Treating research assistance as a starting point to verify, not an authoritative source.

5.4

Appropriate Use

Determining when AI is suitable and when human expertise, conventional tools, or authoritative sources should lead, based on task and stakes.

5.5

The Verification Habit

Verifying facts, figures, dates, names, quotes, and citations before acting on or sharing output.
Understanding that the user, not the AI, remains accountable.

5.6

Personal Prompt Library

Awareness level

The purpose of saving and organising reusable prompts and templates for recurring tasks. (awareness)

Domain 6

Limitations, Risks & Hallucination

16% · 19Q

Critical evaluation and verification as a core professional competency.

6.1

Hallucination

What hallucination is: confident, fluent output that is factually wrong, with no signal that anything is amiss.
Why it is inherent to how language models work, rather than an occasional bug.

6.2

Stale Information & Fabrication

Out-of-date output arising from the knowledge cut-off.
Fabricated statistics, citations, quotes, and case studies that appear authoritative.

6.3

Provenance & Citation Checking

That generated content is not the same as sourced content, and that citations, statistics, and quotes are high-risk.
Confirming important information at its original source.

6.4

Reducing the Risk

Grounding prompts in real content, asking the model to flag uncertainty, cross-checking against primary sources (not another AI), and keeping a human in the loop.

6.5

Mandatory-Verification Contexts

Recognising contexts where verification is required: legal, medical, financial, safety, HR/people decisions, and anything published externally.

Domain 7

AI Security, Privacy & Data

10% · 12Q

Protecting information and recognising AI-enabled risks at a professional-literacy level.

7.1

Protecting Information

Data that must not be entered into public or unapproved AI tools: personal data, confidential and commercially sensitive information, credentials, and confidential documents.
The default rule: "when in doubt, leave it out."

7.2

Data Handling by AI Services

That third-party services may transmit, store, or use inputs according to their terms.
Basic concepts of data retention and residency, and that organisational policy governs use.

7.3

Shadow AI

Shadow AI as unapproved AI use outside organisational oversight, and the risks it creates.
Using approved tools and seeking approval for new ones; treating a policy gap as a reason to escalate.

7.4

AI-Enabled Threats

Recognising deepfakes, impersonation, and AI-assisted phishing and social engineering.
Verifying unexpected or urgent requests through a separate, trusted channel. (awareness)

7.5

Prompt Injection

Awareness level

Prompt injection as malicious instructions hidden in content the AI processes; caution with untrusted content and oversight of AI actions. (awareness)

Domain 8

Ethics, Bias, Responsible AI & Governance

8% · 10Q

Responsible use, oversight, disclosure, and a professional decision framework.

8.1

Bias & Fairness

Sources of bias: training data, design choices, and deployment context.
Real-world impacts (e.g. hiring, lending, content), and bias assessment as a duty in people-affecting uses.

8.2

Transparency, Accountability & Explainability

That the human or organisation remains accountable — "the AI did it" is not a defence.
Transparency and explainability at Foundation level.

8.3

Human Oversight

Why human oversight is necessary, and where automated decision-making increases risk.
The principle that AI proposes while a person decides and owns the outcome.

8.4

Disclosure

When to disclose AI use: where required, where AI substantially generated the content, or where non-disclosure would mislead.

8.5

Policy & Governance

The purpose of an organisational AI policy and basic governance (oversight, accountability, review).
The high-level, risk-based regulatory picture. (awareness)

8.6

Professional Decision Framework

Applying the framework: use / use with verification / escalate / do not use — being more cautious when unsure.
Demonstrating the central principle that AI can assist while humans remain accountable.