Automation Anywhere Under the EU AI Act: Classify the Process, Not the Bot

For most companies, Automation Anywhere is minimal-risk under the EU AI Act. Moving invoices between systems, reconciling spreadsheets, or routing a service ticket is automation over your own operations — beyond the Article 4 AI-literacy duty that attaches to any AI system you deploy, there is no mandatory obligation under Regulation (EU) 2024/1689 for any of it.

The classification does not attach to the product. It attaches to the process each bot or AI agent runs. A bot that copies data between two finance systems and an AI agent that screens job applicants are the same software with two very different legal weights. The first is minimal-risk; the second sits in Annex III and can be high-risk. You classify the work, not the vendor.

This page sets the default, names the precise conditions that flip a specific automation into high-risk territory, and shows where the GDPR layer applies regardless of the AI Act answer.

---

What Automation Anywhere actually does

Automation Anywhere is a robotic process automation (RPA) and agentic-automation product. Its older heritage is rules-based RPA — software bots that mimic clicks, keystrokes and data entry across applications. Its current Agentic Process Automation (APA) line goes further: AI Agent Studio lets teams build goal-driven AI agents, and a Process Reasoning Engine coordinates agents, bots, APIs and documents, with human-in-the-loop checkpoints, to run multi-step workflows that adapt when inputs change.

Two layers matter for classification:

• Deterministic RPA bots that follow fixed rules — open a file, read a field, paste it elsewhere — are not even "AI systems" in many deployments. A bot that just executes pre-set steps with no inference is automation, not an AI system under Article 3(1), and falls outside the Act entirely.
• AI agents built in AI Agent Studio use general-purpose AI (GPAI) models to reason over messy inputs and decide next actions. These are AI systems. But "is an AI system" is not "is high-risk" — that depends on what the agent is used for.

Underneath the AI agents sit GPAI models. Those models carry their own duties under Chapter V (Article 53) — technical documentation, downstream information, a copyright policy, a training-data summary — but those obligations rest with the model provider, not with you as a deployer of Automation Anywhere.

The default and the flip

The decisive question is never "is Automation Anywhere high-risk?" It is: what process does this specific bot or agent automate, and is that process an Annex III use case?

Under Article 6 and Annex III, an AI system is high-risk when it is used in one of the eight listed areas. The two that catch automated workflows most often are:

• Annex III point 4 — employment, workers management and access to self-employment: AI used to recruit or select people, to make decisions on promotion or termination, to allocate tasks, or to monitor and evaluate performance and behaviour.
• Annex III point 5 — access to essential private and public services: including, at point 5(b), evaluating the creditworthiness of natural persons or establishing their credit score (fraud detection excluded).

If you wire an Automation Anywhere agent to read CVs and shortlist candidates, that deployment is an employment-decision system and is potentially high-risk — not because the RPA tool is a hiring product, but because you put it to a hiring use. The same software running payroll reconciliation stays minimal-risk.

Two guardrails keep teams honest:

• A bot that executes a human's decision is not the same as a bot that makes it. Automating the paperwork after a manager has chosen who to hire is administration. Automating the choice is an Annex III assessment.
• The Article 6(3) filter exists, but is narrow. An Annex III system is not high-risk where it performs a purely narrow procedural task, or merely supports a human assessment without replacing or unduly influencing it. A bot that flags incomplete application forms for a recruiter may qualify; a bot that ranks candidates does not. You must document the assessment — and any system that profiles natural persons is always high-risk regardless of the filter.

The GDPR layer comes first

Before the EU AI Act question even arises, an Automation Anywhere bot that touches personal data raises a data-protection one — and that bites even where the AI Act answer is "minimal".

• Lawful basis and purpose limitation (GDPR Articles 5–6). Automating the processing of customer, employee or applicant records still needs a lawful basis and must stay within the original purpose.
• Solely automated decisions (GDPR Article 22). If a bot produces a decision with legal or similarly significant effects on a person without meaningful human involvement, GDPR Article 22 restrictions apply — separate from, and additional to, any AI Act duty. An end-to-end automated rejection of a benefit or a loan is the textbook trigger.
• DPIA (GDPR Article 35). High-risk processing may require a Data Protection Impact Assessment under GDPR — distinct from the AI Act's Fundamental Rights Impact Assessment (Article 27).

For ordinary back-office automation these duties are routine. They sharpen exactly where the AI Act flip also looms — when a bot starts shaping outcomes for identifiable people.

Worked example: Rhenania Services GmbH, a 220-person BPO

Rhenania Services GmbH, a German business-process outsourcer, ~220 employees, ~€31M turnover, runs Automation Anywhere across client accounts. Three automations, three answers.

Bot one: invoice processing. A rules-based bot reads supplier invoices, matches them to purchase orders and posts them to the ledger. No inference, no decision about a person. This is deterministic RPA — arguably not even an AI system, and in any case minimal-risk.

Agent two: ticket triage. An AI agent reads inbound support emails and routes them to the right queue with a suggested priority. It reasons over text but assesses tickets, not people, and a human agent handles every case. Minimal-risk under the AI Act; the GDPR duties on the message content still apply.

Agent three: candidate shortlisting for a client. For a recruitment client, Rhenania builds an agent that reads applications and produces a ranked shortlist that the client's recruiters act on. This sits in Annex III point 4 — employment. Because the output materially drives a selection decision about people, and ranking applicants profiles them, the deployment is high-risk: the Article 6(3) filter cannot rescue a system that profiles natural persons, however closely the recruiters review the shortlist. Rhenania, running the agent under its own authority, is the deployer (Article 26); if it built and supplied the shortlisting agent to the client under its own name, Article 25 could make it the provider with the full Article 16 stack, due by 2 December 2027 under the Digital Omnibus — Regulation (EU) 2026/1744, in force since 27 July 2026, which moved the original 2 August 2026 date.

Same product, same company — only the third automation created a high-risk programme. The difference was the process each bot was pointed at, not the procurement.

On exposure: a deployer breach of the high-risk requirements sits at €15,000,000 or 3% of worldwide annual turnover under Article 99(4), and for an SME the Article 99(6) cap applies the lower of the two figures.

How Confir helps

Confir registers each automation you build in Automation Anywhere as a distinct AI system and classifies it by the actual process it runs under Articles 5 and 6 with Annex III logic — so the invoice bot, the ticket-triage agent and the candidate-shortlisting agent get three separate verdicts, and the Article 6(3) filter and the GDPR cross-map are applied explicitly. The engine is deterministic and rule-based: same intake, same finding, with the rule that fired shown in plain English. Where an automation reaches high-risk, it scopes the Article 26 deployer duties and the Article 27 FRIA where relevant.

Frequently Asked Questions

Is Automation Anywhere high-risk under the EU AI Act?

By default, no. Most Automation Anywhere bots automate back-office work — moving data, reconciling records, routing tickets — which touches no Annex III use case and carries no mandatory obligations beyond Article 4 AI literacy for the AI-driven parts. A specific automation becomes potentially high-risk only when the process it runs is an Annex III decision about a person, such as screening job applicants (point 4) or scoring creditworthiness (point 5(b)). You classify the process, not the product.

Are rules-based RPA bots even AI systems under the Act?

Often not. A deterministic bot that follows fixed steps with no inference — open a file, copy a field, paste it elsewhere — may not meet the Article 3(1) definition of an AI system at all, and so falls outside the Act. The agentic features built in AI Agent Studio, which reason over inputs using GPAI models, are AI systems — but being an AI system is not the same as being high-risk.

When does an Automation Anywhere agent become high-risk?

When you point it at an Annex III process. The common triggers are employment decisions under Annex III point 4 — recruiting, selecting, evaluating or allocating work to people — and access to essential services under point 5, including creditworthiness scoring at point 5(b). Automating the admin around a human decision stays minimal-risk; automating the decision itself crosses the line.

Does the Article 6(3) exemption help if a human reviews the output?

It can. Article 6(3) means an Annex III system is not high-risk where it performs a narrow procedural task or merely supports a human who decides on their own judgment, without being unduly influenced. A bot that flags incomplete forms may qualify; a bot that ranks candidates does not. You must document the assessment, and any system that profiles natural persons is always high-risk regardless of the filter.

Do GDPR obligations apply to Automation Anywhere bots?

Yes, and often before the AI Act question arises. Where a bot processes personal data you need a lawful basis (GDPR Articles 5–6) and must respect purpose limitation. If a bot makes a solely automated decision with significant effects on a person, GDPR Article 22 applies, and a DPIA under GDPR Article 35 may be required — distinct from the AI Act's Article 27 FRIA. Treat the GDPR check as the first gate.

Are we a deployer or a provider when we use Automation Anywhere?

Almost always a deployer (Article 26) — you run bots under your own authority for your own purposes. You become a provider under Article 25 only if you place an AI system you built on the product on the market under your own name, substantially modify a high-risk system, or repurpose one for a high-risk use. For most internal automation, deployer status is the right answer, and the deployer duties are lighter than the provider stack.

Related guides

• Salesforce Einstein under the EU AI Act
• ServiceNow AI under the EU AI Act
• Annex III — the eight high-risk use cases explained
• Article 6 — how high-risk AI systems are classified
• Article 26 — deployer obligations explained
• Deployer — definition and obligations
• Provider vs deployer — which one are you?