Regulatory regime

Last updated Oct 6, 2026

EU AI Act

An Ethotechnics critique of the EU AI Act: most high-risk systems are self-assessed, their authority does not lapse, and a challenge need not reach them.

Ethotechnics critique

Where the regime falls short

The Act requires assessment before release and monitoring after it (Art. 72). The gaps are in what the running system owes the people it decides about.

  • Self-assessed authority does not expire. Most high-risk systems listed in Annex III are assessed by their own provider before release (Art. 43(2)), and assessed again only after a substantial modification (Art. 43(4)). A notified body's certificate lapses after at most four or five years (Art. 44(2)); a provider's own assessment does not, and nothing requires it to be renewed on evidence.
  • A challenge does not have to reach the system. A person affected by a high-risk system's decision can ask the deployer to explain it (Art. 86), and anyone can complain to a market surveillance authority that the Act was breached (Art. 85). The Act gives no route to contest the decision itself; that comes from other law, such as GDPR Article 22. No provision in either ties an upheld challenge to a change in the system.
  • Cost to people is assessed before use, not counted after. Public bodies, private entities providing public services, and deployers that score credit or price life and health insurance assess the impact on fundamental rights before first use, and update the assessment when its elements change (Art. 27). Nothing requires a measure of what the running system costs the people it decides about, such as the time they spend correcting its errors.
  • A halt is required; rehearsing it is not. Human oversight must, as appropriate and proportionate, let a person interrupt a high-risk system so it halts in a safe state (Art. 14(4)(e)). Nothing requires anyone to rehearse the halt or meet a recovery deadline.

Failure mode

Lawful deployment + harmful latency

Compliance can coexist with prolonged pending states and slow restoration.

The failure mode is a system that is compliant yet still traps people in extended “pending” states without enforceable restoration timelines.

Application dates

When each part applies

Article 113, as amended by Regulation (EU) 2026/1744, the Digital Omnibus on AI, in force since July 27, 2026.

Date What applies
Entry into force.
Chapters I and II: general provisions, including AI literacy (Art. 4), and the prohibited practices in Art. 5.
Obligations for general-purpose AI models (Chapter V), governance (Chapter VII), penalties (Chapter XII, except Art. 101), and notified bodies (Chapter III, Section 4).
The general application date: most remaining provisions, including the transparency obligations in Art. 50.
The prohibitions the omnibus added: AI systems that generate non-consensual intimate imagery or child sexual abuse material. The Art. 50(2) transition ends for certain providers of systems that generate synthetic content and were on the market before August 2, 2026.
Chapter III, Sections 1 to 3, for high-risk systems under Art. 6(2) and Annex III. The original date was August 2, 2026.
Chapter III, Sections 1 to 3, for high-risk systems under Art. 6(1) and Annex I, AI in products covered by EU harmonisation legislation. The original date was August 2, 2027.

Checked on October 6, 2026 against the Commission's AI Act Service Desk timeline, which reflects the omnibus amendments.

Articles to clauses

From the articles to the standards

The framework map takes Articles 9, 12, 14, 72, and 73 one at a time: the Ethotechnics clauses that bear on each, and what they leave to you.

No clause covers serious-incident reporting under Article 73. Read the article table.

Implications

What organizations should optimize for right now

Prepare for converging governance expectations, not a single date bet.

  • Prioritize auditability primitives: receipts, decision records, evidence packs, and repair logs that survive policy edits.
  • Operationalize time bounds: define and test time-to-human, time-to-remedy, and escalation paths with named authority.
  • Keep decisions open to challenge: include standing, non-retaliation, and reliable routes to challenge that can be verified in an audit.
  • Crosswalk controls: map one control library across the EU AI Act, NIST AI RMF, and ISO/IEC 42001 so readiness does not hinge on one regime.

Maintenance

Sources and updates

  • Sources: Article 113 of Regulation (EU) 2024/1689 as amended by Regulation (EU) 2026/1744, and the Commission's AI Act Service Desk timeline.
  • Updates: the page changes when a date, an obligation's scope, or an enforcement mechanism changes. The date it was last checked is at the top of the page.
Cite this page · APA, MLA, Chicago, BibTeX, RIS

Version

vDraft

Last updated

Oct 6, 2026

APA

Ethotechnics Institute. (2026). EU AI Act. Ethotechnics Institute. https://ethotechnics.org/standards/critiques/eu-ai-act

MLA

Ethotechnics Institute. "EU AI Act." Ethotechnics Institute, 2026, https://ethotechnics.org/standards/critiques/eu-ai-act.

Chicago

Ethotechnics Institute. "EU AI Act." Ethotechnics Institute. Oct 6, 2026. https://ethotechnics.org/standards/critiques/eu-ai-act.

BibTeX

@misc{ethotechnics_standards_critiques_eu_ai_act,
  title={EU AI Act},
  author={Ethotechnics Institute},
  year={2026},
  howpublished={Ethotechnics Institute},
  url={https://ethotechnics.org/standards/critiques/eu-ai-act},
  version={Draft}
}

RIS

TY  - WEB
TI  - EU AI Act
AU  - Ethotechnics Institute
PY  - 2026
UR  - https://ethotechnics.org/standards/critiques/eu-ai-act
ER  -