monday.com AI and the EU AI Act: A Work OS, Classified by Use

Minimal risk for almost every deployment — but monday.com is configurable enough that the answer depends on what you build. monday.com AI delivers its intelligence through AI blocks that slot into columns, automations and the workflow builder, plus native AI agents teams can configure and deploy. Used to summarise updates, categorise items, draft text and automate routine board work, that activity matches none of the high-risk purposes in Annex III. It sits at minimal risk. Because monday is a work operating system rather than a fixed-purpose app, the classification follows the workflow you assemble, not the brand on the login screen.

This guide is for the deployer. The provider of the underlying general-purpose AI model carries the Chapter V obligations (Articles 51–55, in force since 2 August 2025); your duties are light — Article 4 AI literacy, a narrow slice of Article 50 transparency, and governance of where your AI blocks and agents actually run. The configurability is the thing to watch: an AI block built onto the wrong board is the one realistic path to a heavier tier.

---

Classify the workflow, not the work OS

The right question is not "is monday.com AI high-risk?" but "is this workflow I built high-risk?" For project tracking, CRM-style pipelines, marketing calendars and ops boards, the answer is no. Annex III, read with Article 6(2), lists the high-stakes purposes that make a stand-alone system high-risk: employment and worker management, access to essential services, creditworthiness, education, critical infrastructure, biometrics. An AI block that summarises a project update or assigns a task by workload is none of those.

Because monday lets you compose AI columns, automations and agents freely, two companies can run the same product at two different tiers. The product is minimal-risk by default; a specific workflow you assemble — say, an AI agent that scores job applicants on a recruitment board — is what could change the answer. The heavy model obligations, meanwhile, sit upstream with whoever trained the foundation model behind the AI blocks. Under Articles 51–55 that provider holds the technical documentation, the training-data summary and the copyright-policy duties; you benefit without inheriting them. Role determines obligation, and yours is deployer (Article 26).

---

What actually triggers a duty — and what does not

Minimal risk: everyday board automation

AI blocks that summarise items, categorise or tag entries, extract data, draft text and automate routine board moves over your own operational data carry no conformity assessment, no Article 11 technical documentation, no Annex IV file, and no Article 49 registration. This is the bulk of how teams use monday.com AI, and it attracts no mandatory AI Act obligation beyond literacy.

Limited risk: when an agent faces outsiders (Article 50)

Article 50 transparency bites where AI output crosses to people outside your organisation. A monday AI agent that answers external customers brings Article 50(1) into play: people must be told they are dealing with an AI system unless it is obvious. That design duty sits with monday as provider — check the notice shows in your channel, and treat it as yours if you put the agent out under your own brand. AI-drafted emails sent by an automation are lighter still: Article 50(4) reaches only deepfakes and text published to inform the public on matters of public interest. Internal board automation never reaches a third party, so for most monday AI footprints Article 50 does not apply.

The one path to high-risk — what you assemble on an HR board

monday is widely used to run recruitment pipelines, onboarding, performance reviews and headcount planning. The moment an AI block or agent on one of those boards produces a decision about people — ranking candidates, screening applicants out, scoring employees for promotion, evaluation or task allocation that sets someone's terms — you are in Annex III point 4, employment and worker management. Point 4(a) covers recruitment and selection; point 4(b) covers decisions about people already in work.

This is a build choice, not how monday.com AI behaves out of the box. An AI block that tags an incoming application by region is administrative; an AI agent that ranks applicants into a shortlist is a recruitment decision system. Even where an Annex III use is in play, the Article 6(3) filter can take a narrow procedural use back out of scope — for instance where the AI only improves a completed human assessment without influencing the outcome — but any system that profiles a natural person is always high-risk, which keeps that exit narrow on a people-facing board.

---

The three realistic duties for a monday.com AI deployer

Article 4: AI literacy for the people who build and use the blocks

Article 4 has applied since 2 February 2025 and binds you whatever the risk tier. The people configuring AI columns and agents especially need to understand what an AI block can and cannot reliably do — that a summary omits, that a categorisation can be wrong, that an AI-drafted message needs a human check before it leaves the building. A short acceptable-use note plus role-appropriate guidance satisfies this; no formal programme is required at minimal risk.

Article 50: disclosure where an agent or automation faces outsiders

Check Article 50 only where monday AI output reaches third parties — chiefly an interacting-with-AI notice on any customer-facing agent, which monday as provider must build in and you should verify. Article 50 has applied in full since 2 August 2026; machine-readable marking of generated content is likewise monday's duty, and your own Article 50(4) disclosure covers only deepfakes and public-interest text. For purely internal automation the obligation is empty, because nothing external is produced.

Governing what gets built: the duty the Act assumes you have

monday's strength — anyone can assemble an AI block or deploy an agent without code — is also where ungoverned high-risk use can appear. The Act assumes you know what AI you run and on what data. List monday.com AI in your inventory, record which boards and data categories AI blocks and agents may touch, and set a rule that people-decision boards get a classification check before an AI block goes live. That single control is what stops a well-meaning HR automation from quietly becoming an unassessed Annex III system.

One data-governance point sits alongside these: GDPR applies independently of the AI Act, so confirm a lawful basis and a data-processing agreement before personal or special-category data flows into AI columns or agent prompts.

---

monday.com AI obligation map by use case

The table maps common patterns to risk tier, the duty each triggers, and the reference. Minimal and limited dominate; only the deliberately built employment rows reach high-risk.

The tier column carries the message: minimal by default, high-risk only on a board you deliberately point at people-decisions.

---

Worked example: a 160-person scale-up runs monday across the company

Talvio OÜ, a 160-person logistics-tech scale-up in Tallinn with around €24M turnover, runs monday.com across operations, marketing and support. AI blocks summarise delivery-exception items, categorise inbound tickets and draft internal status notes; one AI agent answers routine questions from external customers in a support portal.

The operations and marketing automation is minimal risk; Article 4 literacy is the only duty. The external support agent brings in Article 50(1) — customers need a clear notice that they are talking to an AI agent, which monday must build in and Talvio checks is visible in its portal. No Annex III purpose is in play, so no high-risk obligation attaches. The action list is one inventory entry, a one-page literacy note, an interaction notice on the support agent, and a data rule keeping customer special-category data out of agent prompts (GDPR Article 9 layer).

Now the counterfactual. Talvio's people team later builds a recruitment board where a monday AI agent scores and ranks applicants and auto-rejects the bottom tier. That single workflow becomes Annex III point 4(a) high-risk from 2 December 2027, pulling in Article 26 deployer duties — human oversight, log retention, informing affected candidates — and, because Talvio itself turned a general agent into a candidate-selection system, a check on whether Article 25(1)(c) makes it the provider. No Article 27 FRIA is due: a private employer outside public services is not among the deployers Article 27 covers. It is a new, deliberate system, not the rollout described, which stays a half-day inventory-and-policy exercise.

---

Deadlines and penalties — proportionate, not alarmist

The settled dates: Article 4 AI literacy and the Article 5 prohibitions apply since 2 February 2025; GPAI obligations (Articles 51–55) since 2 August 2025; Article 50 transparency in full since 2 August 2026, with a 2 December 2026 marking deadline for generative systems already on the market before that date.

High-risk timing matters here only as a contingency. The Digital Omnibus — Regulation (EU) 2026/1744, published in the Official Journal on 24 July 2026 and in force since 27 July 2026 — moved the stand-alone Annex III high-risk deadline from 2 August 2026 to 2 December 2027. The deferral is a fixed calendar date, not the standards-contingent "stop the clock" version that was rejected.

On penalties: Article 50 and most deployer-obligation breaches carry fines up to €15 million or 3% of total worldwide annual turnover, whichever is higher (Article 99(4)). Prohibited-practice breaches under Article 5 reach €35 million or 7% (Article 99(3)). Supplying incorrect or misleading information to authorities carries up to €7.5 million or 1% (Article 99(5)). For SMEs and start-ups, each fine is capped at the lower of the percentage or the fixed amount (Article 99(6)). For minimal-risk board automation there is no fine exposure beyond getting the light disclosure right.

---

How Confir helps

List monday.com AI in Confir's AI register and answer the intake questions about role, use case, data types and audience — and, because monday is configurable, classify the specific workflow rather than the product as a whole. The deterministic, rule-based engine — no model inference, no hallucination — derives the tier (minimal, the Article 50 limited slice, or the Annex III employment path) and the matching obligation checklist, stating in plain language the exact rule that fired. The output is a dated, reproducible record of why each AI workflow was classified the way it was — and a flag the moment someone points an AI block at a people-decision.

Related guides

• build your AI system inventory
• detect shadow AI across your workspace
• what minimal risk means under the Act
• is recruitment-screening AI high-risk?
• employment-decision AI and Annex III
• Article 50 transparency obligations