Unadhi Guides

"Is it an AI system?" check sheet for accountants: sort a small business client's software list under Article 3(1) of the EU AI Act and the Commission guidelines C(2025) 5053

Updated 29 Oct 2026 · Disclosure: the EURO AI Register mentioned below is an Unadhi product with a free edition. This guide is for information only, not legal advice.

What you get: a finished "Is it an AI system?" check sheet for accountants, tax advisers and bookkeepers to run over a small business client's software list. One row per tool or feature; you answer six short questions and the sheet gives a result (AI SYSTEM: ADD TO REGISTER, LIKELY NOT AN AI SYSTEM, ASK VENDOR or CHECK BY HAND) plus the reason with the paragraph of the Commission's guidelines behind it. A tab of questions to send software vendors, a "what follows" checklist and a client email follow. Made-up example rows show how it fills in.

Short answer: the EU AI Act only applies to software that meets the definition of an "AI system" in Article 3(1) of Regulation (EU) 2024/1689. The key test is whether the system infers how to generate outputs (predictions, content, recommendations or decisions), using machine learning or logic- and knowledge-based techniques (Recital 12). Software that only follows rules written by people, basic data processing such as standard spreadsheets and sorting or filtering, descriptive statistics and charts, classical heuristics, and simple prediction such as "next month = the average" is outside the definition according to the Commission's guidelines (C(2025) 5053, paragraphs 40 to 51). The guidelines are not binding, give no automatic list and say each system must be assessed on its own features (paragraphs 6, 7 and 62).

Why clients ask their accountant this question

The seven parts of the definition (Article 3(1))

"'AI system' means a machine-based system that is designed to operate with varying levels of autonomy and that may exhibit adaptiveness after deployment, and that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments." The guidelines split this into seven elements (paragraph 9). For a client software list, three do the work:

ElementWhat it means in practiceSource
Infers how to generate outputsBuilt with machine learning (supervised, unsupervised, self-supervised, reinforcement, deep learning; language models are given as an example) or logic- and knowledge-based approaches that reason from encoded knowledge (expert systems, inference engines). This is the "key, indispensable condition".Recital 12; guidelines paras 26, 30-39
Some autonomyThe system produces an output without a person manually specifying it. A system that needs manual inputs but then produces the output itself still counts. Only systems with full manual human involvement are excluded.Recital 12; paras 17-20
OutputsPredictions, content, recommendations or decisions. A recommendation applied automatically becomes a decision.paras 52-58
AdaptivenessSelf-learning after deployment is not required ("may"). A tool that never changes can still be an AI system.paras 22-23

Step 1: tab 1, software list (one row per tool or feature)

ColumnWhat to writeSource
ATool or feature (list AI features of a bigger product separately)Client's software list
BVendor or productInvoice or app store
CWhat it does, in one lineVendor documentation
DBuilt with machine learning, including a language model or other trained model? Y, N or ?Recital 12; paras 32-38
EReasons from encoded expert knowledge (expert system, inference engine), beyond fixed if-then steps? Y, N or ?para 39
FOnly rules, formulas, filters or macros written by people? Y or NRecital 12; para 46
GOnly basic statistics: totals, averages, charts, a "mean" baseline forecast, or long-established optimisation such as linear or logistic regression? Y or Nparas 42, 47, 49-51
HOutput: type exactly Prediction, Content, Recommendation, Decision or Noneparas 52-58
IProduces its output without a person specifying it each time? Y or Nparas 17-18
JResult: =IF(A2="","",IF(OR(D2="?",E2="?"),"ASK VENDOR",IF(OR(D2="Y",E2="Y"),IF(OR(G2="Y",H2="None",I2="N"),"CHECK BY HAND","AI SYSTEM: ADD TO REGISTER"),IF(OR(F2="Y",G2="Y"),"LIKELY NOT AN AI SYSTEM","CHECK BY HAND"))))Formula (our working rule)
KReason: =TEXTJOIN("; ",TRUE,IF(D2="Y","learns from data (paras 32-38)",""),IF(E2="Y","infers from encoded knowledge (para 39)",""),IF(OR(D2="?",E2="?"),"technique unknown: send the vendor questions",""),IF(F2="Y","only rules written by people (Recital 12, para 46)",""),IF(G2="Y","only basic statistics or simple prediction (paras 42, 47, 49-51)",""),IF(H2="None","no prediction, content, recommendation or decision (paras 52-58)",""),IF(I2="N","every output set by a person (paras 17-18)",""))Formula
LEvidence (link to vendor page or reply) and date checkedYour file

The result in column J is our working rule for sorting a list quickly, not a legal test. "CHECK BY HAND" means the answers point both ways (for example, a trained model whose vendor says it is plain linear regression); read guidelines paragraphs 40 to 51 for that row and write down your reasoning in column L.

Example data (made up, to show how the sheet fills in; no real products):

ACDEFGHIJK
Example 1: fee spreadsheetSUMIF and lookup formulas, no AI features switched onNNYNNoneNLIKELY NOT AN AI SYSTEMonly rules written by people (Recital 12, para 46); no prediction, content, recommendation or decision (paras 52-58); every output set by a person (paras 17-18)
Example 2: bank feed auto-categorisationSuggests an account code per bank line, learns from correctionsYNNNRecommendationYAI SYSTEM: ADD TO REGISTERlearns from data (paras 32-38)
Example 3: invoice data captureReads supplier invoices into the ledger; vendor page only says "smart"??ContentYASK VENDORtechnique unknown: send the vendor questions
Example 4: website FAQ chatbotAnswers visitor questions in its own wordsYNNNContentYAI SYSTEM: ADD TO REGISTERlearns from data (paras 32-38)
Example 5: sales reportTotals, averages and trend chartsNNNYNoneNLIKELY NOT AN AI SYSTEMonly basic statistics or simple prediction (paras 42, 47, 49-51); no prediction, content, recommendation or decision (paras 52-58); every output set by a person (paras 17-18)
Example 6: cash-flow forecastNext month = average of the last 12 monthsNNYYPredictionYLIKELY NOT AN AI SYSTEMonly rules written by people (Recital 12, para 46); only basic statistics or simple prediction (paras 42, 47, 49-51)
Example 7: stock reorder toolVendor says: linear regression on past sales, unchanged for yearsYNNYRecommendationYCHECK BY HANDlearns from data (paras 32-38); only basic statistics or simple prediction (paras 42, 47, 49-51)
Example 8: email ruleMoves mail with "invoice" in the subject to a folderNNYNNoneYLIKELY NOT AN AI SYSTEMonly rules written by people (Recital 12, para 46); no prediction, content, recommendation or decision (paras 52-58)

We tested both formulas in LibreOffice Calc on these rows on 29 Oct 2026. In older Excel without TEXTJOIN, use the column J result alone.

Step 2: tab 2, questions for the software vendor (for every "ASK VENDOR" row)

  1. Does this feature use a machine learning model, including a large language model, trained on data? If so, whose model is it?
  2. If it is rule-based: were all the rules written by people, or does the system derive rules or conclusions itself from a knowledge base?
  3. What does it produce: predictions, generated text or images, recommendations, or decisions?
  4. Is any output applied automatically, without a person reviewing it?
  5. Does it change its behaviour while in use, for example by learning from our corrections?
  6. Do you see yourselves as the provider of an AI system under Regulation (EU) 2024/1689 for this feature? Where is that documented?

Paste the reply or link into column L and change D and E from ? to Y or N.

Checklist: what follows for each "AI SYSTEM" row

Email to small business clients (template)

Subject: Which of your tools count as "AI" under the EU AI Act?

Hello [name],

Many software tools now call themselves "AI". Under the EU AI Act only some of them are "AI systems", and that is the starting point for the AI literacy duty and for any further checks. Please send us by [date] a list of the software and add-ons your business pays for or uses at work (accounting, invoicing, CRM, email, website chat, scanning, scheduling, HR), and tell us which features you actually switched on. We will sort each one into "AI system", "likely not" or "ask the vendor" using the Commission's guidelines, and send you the short list of vendor questions where we cannot tell. This is a working list, not legal advice.

Thank you,
[your name]

Related free sheets: client AI inventory for accountants · provider or deployer role check · Article 4 AI literacy record · Annex III high-risk check · Article 5 prohibited practices list.

Keep the AI rows in a register

Copy every "AI SYSTEM: ADD TO REGISTER" row into the EURO AI Register free edition. It runs in your browser with no account and no upload, covers your first 10 AI systems and 10 evidence records, the screening questions, duties by role with dates, and a printable Article 4 AI-literacy record. Attach this sheet as the evidence record for the "is it AI?" decision.

When a client has more than 10 systems, the EURO AI Register membership starts at $19 a month and adds unlimited systems, rule updates and email help. Neither tool is legal advice or proof of compliance.

Partner programme opening soon

Partner programme opening soon: join the free waitlist. Planned terms are 50% of each referred sale (75% once you have 5+ paying customers), single level, paid through Gumroad only from revenue received.

Limits and open points

The Commission's guidelines are not binding; only the Court of Justice of the European Union can give an authoritative interpretation (paragraph 7), and the Commission says no automatic determination or exhaustive list is possible (paragraph 62). The guidelines do not cover general-purpose AI models (paragraph 64). Article 3(1) itself was not changed by Regulation (EU) 2026/1744, but several other articles were, so read the current text on EUR-Lex. This sheet covers the AI Act only: data protection, consumer and sector rules apply separately. When a row matters, for example hiring or credit decisions, get advice.

Frequently asked questions

Is an Excel spreadsheet an AI system under the AI Act?

A standard spreadsheet without AI-enabled features is given by the Commission as an example of basic data processing outside the definition (guidelines paragraph 46). AI features switched on inside a spreadsheet are a separate row to check.

Does a tool have to keep learning to be an AI system?

No. Article 3(1) says the system "may" exhibit adaptiveness after deployment; the Commission says this is not a decisive condition (paragraphs 22 and 23). The decisive condition is the capability to infer (paragraph 26).

Is a chatbot built on a language model an AI system?

The guidelines give language models that learn to predict the next token as an example of self-supervised machine learning, and content generated with GPT-type models as an AI output (paragraphs 36 and 56). Using one at work normally makes the business a deployer.

Not legal advice. The sheet helps sort a software list; it does not decide whether the AI Act applies to a business. Spotted an error? Write to mininglife7@gmail.com and we correct it.

Sources

← All guides

Unadhi Guides are free, independent explainers. We correct errors as soon as we find them: write to mininglife7@gmail.com.