Zum Hauptinhalt springen
ISO/IEC 27701:2025 · Annex A, Table A.1 · Control for PII controllers

A.1.3.11 Automated decision making

Updated on 5 min Reviewed by: Cenedril Editorial
ISO/IEC 27701:2025GDPR

A recruitment agency uses a platform that scores CVs and automatically rejects applications below a threshold. The rejection goes out without anyone looking at it. One rejected applicant asks why she was screened out and requests that a human look at her documents. Only now does the organization realize that it has been making automated individual decisions for months, with no information, no review route and no impact assessment. A.1.3.11 requires the organization to identify every decision about people that results solely from automated processing without human review, and to meet the resulting obligations to data subjects, legal ones included.

As AI tools spread, the number of such decisions grows quickly. They often arise in procurement and business units, where nobody thinks of privacy when a tool “just pre-sorts”.

What does the standard require?

The control closes the group of obligations to PII principals and singles out a special case.

  • Identify and meet obligations. If the organization makes decisions about a person solely on the basis of automated processing, it establishes the resulting obligations, legal ones included, and meets them.
  • Typical obligations. Some jurisdictions attach special duties to decisions with significant effect. The standard names informing people that decisions are automated, the possibility to object to such decision making and access to human intervention.
  • Limits of automation. In a note, the standard points out that in some jurisdictions certain processing may not run fully automated.
  • Plan for compliance. Organizations operating in such jurisdictions take these obligations into account in their procedures.

In practice

Find automated decisions. Flag every activity in the record of processing activities in which a system decides about people. Check purchased software and AI features too, such as those that filter applications, assess credit, set prices or block accounts. The AI register is a good cross-check here.

Settle admissibility. Under the GDPR, solely automated decisions with significant effect are only allowed in certain cases: where they are necessary for a contract, authorized by law or based on the person’s explicit consent. Even tighter limits apply to special categories.

Inform and explain. The privacy notice names the procedure, explains the logic in plain words and describes the consequences for the person. A description like “we use algorithms” does not suffice.

Open the route to a human. Data subjects can state their point of view, contest the decision and request review by a human. The reviewer is trained, has access to all relevant information and may change the decision.

Document and monitor the logic. Record which data feeds in, which criteria apply and how the procedure is tested, including for errors and bias. Such processing almost always needs an impact assessment under A.1.2.6.

Typical audit evidence

For A.1.3.11, auditors typically expect:

  • Flags for automated decisions in the record of processing activities, checked against the AI Register
  • Admissibility check per procedure, naming the exception relied on
  • Information for data subjects on logic, significance and consequences
  • Procedure for human review with logs of the cases handled
  • Impact assessments for the processing concerned (→ Risk Register)

KPI

% of automated decisions with documented logic and human review option

Procedures with automated decisions count if their logic is described and data subjects can request human review. The target is 100%. The figure stands or falls with the completeness of the inventory.

Supporting KPIs:

  • Share of review requests handled by a human within the deadline
  • Share of reviews that led to a changed decision
  • Share of procedures with a completed impact assessment

A.1.3.11 corresponds to Article 22 GDPR. Under paragraph 1, data subjects do not have to accept a solely automated decision, including profiling, that produces legal effects or similarly significantly affects them. Paragraph 2 sets out the exceptions, and paragraph 3 the safeguards: at least the right to human intervention, to express one’s point of view and to contest the decision. Paragraph 4 limits the use of special categories. Articles 13(2)(f), 14(2)(g) and 15(1)(h) require meaningful information about the logic and the significance. Article 35(3)(a) names such procedures as a standard case for an impact assessment. In Germany, section 37 of the Federal Data Protection Act provides an exception for insurance benefits (→ GDPR General Data Protection Regulation). Where decisions rest on high-risk AI systems, the obligations of the AI Act apply in addition (→ EU AI Act: Regulation on Artificial Intelligence).

Automated decisions connect privacy and AI management:

Sources

Frequently asked questions

When is a decision solely automated?

When no human with authority of their own and sufficient information reviews the outcome before it takes effect. A person who routinely confirms the system's recommendations without being able to change them does not turn the result into a human decision. That is the view of the WP 251 guidelines endorsed by the European Data Protection Board.

Is scoring already an automated decision?

It can be. In December 2023, the Court of Justice ruled that calculating a probability value already falls under Article 22 GDPR if a third party draws strongly on it for its decision (C-634/21, SCHUFA). Whoever produces scores for others or uses scores from others therefore checks both sides.

What role does the AI Act play?

It supplements the GDPR. For high-risk AI systems it requires human oversight, and Article 86 gives affected persons a right to an explanation of certain decisions based on such systems. The obligations under Article 22 GDPR continue to apply alongside.

No Cookies!

This wiki collects nothing, bakes nothing and leaves nothing behind. There's nothing to consent to. Privacy doesn't get better than this.