社会ガイド

EU Digital Services Act Rules for Recommender Systems

The EU Digital Services Act (DSA) sets recommender-system transparency duties for covered providers, subject to scope limits in the regulation.

  • 3 分で読めます
  • 最終更新日
このページでは3 分で読めます
  1. 概要
  2. ディープダイブ
  3. 戦略的影響
  4. The Future of EU Digital Services Act Rules for Recommender Systems
  5. 現実世界の実装
  6. リスクとガードレール
  7. 実装ロードマップ
  8. 探検を続けましょう
  9. よくある質問

概要

Article 27 covers providers of online platforms using recommenders, but Article 19 generally excludes micro and small platform providers from Section 3; that section applies to a platform designated as a VLOP regardless of size. Article 38 separately requires VLOPs and VLOSEs using recommenders to provide a non-profiling option for each system.

ディープダイブ

The Digital Services Act (Regulation (EU) 2022/2065) has a layered approach. Article 27 requires providers of online platforms using recommender systems to explain in plain, intelligible language the main parameters used and any options recipients can use to modify or influence them. The explanation must identify the most significant criteria for suggestions and their relative importance. If multiple ordering options exist, a selection control must be directly and easily accessible from the relevant interface section. Article 19 limits this duty’s scope. Micro and small enterprise providers of online platforms are generally excluded from Section 3, where Article 27 appears, except for Article 24(3). Section 3 does apply to a platform designated as a very large online platform (VLOP), regardless of whether its provider qualifies as micro or small. Do not describe Article 27 as covering every small platform without qualification. Article 38 is a separate additional duty: providers of VLOPs and very large online search engines (VLOSEs) that use recommenders must provide at least one option for each system that is not based on GDPR-defined profiling. That duty is not limited to VLOPs; it also names VLOSEs. The law does not prescribe a chronological feed or require public release of source code and model weights. A compliance review should map each product surface to the service provider and its legal designation, identify recommender systems, and check whether explanations and controls match deployed behavior. These core articles do not exhaust the DSA, and details depend on service scope and facts. Check current official material for the provider concerned; this guide is a plain-language overview, not legal advice.

戦略的影響

リスクと安全性

AI による壊滅的な被害も日常的な被害も、誰がリスクを理解し、誰が行動できるかにかかっています。

より明確な判決

国民と専門家のリテラシーは、強力な安全政策が政治的に可能かどうかを左右します。

誇大広告を打ち破る

明確な説明は、誇大広告、研究室の PR、曖昧な倫理劇場に囚われることを減らします。

The Future of EU Digital Services Act Rules for Recommender Systems

As services change their feeds and recommendation features, clear inventories of service scope, ranking parameters, and user controls will make DSA review easier. Commission guidance and enforcement practice may further clarify implementation, so providers should check current official material for their specific service and designation. The durable lesson is to explain the main ranking criteria in usable terms, provide controls where the regulation requires them, and keep the explanations aligned with the system actually operating. Providers should also retain implementation records so that future audits can trace a user-facing explanation to a particular recommender and version.

現実世界の実装

An online platform explains in intelligible terms the principal criteria that determine why content is suggested, consistent with Article 27.

Where a platform offers several ranking options, it makes the preference control directly and easily reachable from the relevant interface section.

A service designated as a VLOP offers a non-profiling option for each of its recommender systems under Article 38.

A compliance reviewer compares terms-of-service explanations and available controls with the deployed interface, while treating this as an audit question rather than assuming the DSA mandates public source code.

リスクとガードレール

  • 能力が複雑になる一方で、実存的なリスクを SF として扱います。

  • 高度な自律性の下での調整による表面製品の安全性を混乱させる。

  • 英語以外や専門家ではない聴衆には、低品質の情報源しか提供されません。

実装ロードマップ

  1. 製品の危害、誤使用、制御不能/調整不良のリスクを分離します。

  2. どのような証拠がタイムラインと重大度についてのあなたの見方を変えるかを尋ねてください。

  3. マーケティング上の主張よりも、一次情報源と具体的な評価を優先します。

  4. 意識だけでなく、キャリア、政策、資金、スキルなど、行動経路を 1 つ特定します。

探検を続けましょう

Free newsletter

Get the daily AI briefing

Three verified AI stories every weekday morning, written in plain English. Free forever, no ads.

One email each weekday. Unsubscribe in one click. We never sell or share your address.

Test yourself

Take the EU Digital Services Act Rules for Recommender Systems quiz

Instant feedback on every answer, and a shareable certificate with a verifiable ID once you pass a course.

クイズを開始する

Support free AI education. AI Understanding is a 501(c)(3) nonprofit — no ads, no paywall, ever. Make a donation

よくある質問

What is EU Digital Services Act Rules for Recommender Systems?

The EU Digital Services Act (DSA) sets recommender-system transparency duties for covered providers, subject to scope limits in the regulation. Article 27 covers providers of online platforms using recommenders, but Article 19 generally excludes micro and small platform providers from Section 3; that section applies to a platform designated as a VLOP regardless of size. Article 38 separately requires VLOPs and VLOSEs using recommenders to provide a non-profiling option for each system.

How does Article 19 affect Article 27 coverage for micro and small platforms?

Article 27 is in Section 3. Article 19 generally excludes micro and small platform providers from that section, but Section 3 applies to a designated VLOP regardless of size.

What must a platform’s Article 27 explanation include about main parameters?

Article 27 calls for the main criteria and why they matter, stated plainly.

When does Article 38 require a non-profiling option for each recommender system?

Article 38 names providers of VLOPs and VLOSEs as the covered providers.

A platform offers several feed-ordering options. Where should the selection function be reachable under Article 27?

Article 27 requires an accessible function in the relevant prioritized-content section.

Does Article 27 require a platform to disclose all source code and model weights?

The guide distinguishes the statute’s user-facing requirements from code disclosure.