The EU AI Act is discussed often, but its timeline is frequently misquoted — especially the claim that "from August 2026 every high-risk AI system must comply". This article checks the dates against the official implementation timeline and points out what actually concerns software companies in Vietnam.
The dates, per the official timeline
- 2 August 2026 — transparency obligations for AI-generated content under Article 50(2): synthetic content must be marked in a machine-readable way. This is the date most often mistaken for a general high-risk deadline.
- 2 December 2027 — Chapter III requirements apply to high-risk systems under Article 6(2) and Annex III (recruitment, education, credit scoring, critical infrastructure and so on).
- 2 August 2028 — the same requirements apply to high-risk systems under Article 6(1) and Annex I, that is, AI embedded in products already covered by sector safety rules.
- 2 August 2030 — high-risk systems used by public authorities must have completed compliance.
In other words: the pressing 2026 date is about transparency, while the heavy obligations for high-risk systems fall in late 2027 and 2028.
Who is affected without being in Europe
The Act applies based on where a system is placed on the market or used, not where the company is registered. Three situations come up frequently for Vietnamese software companies:
- Outsourcing for European clients. The client carries the primary obligation, but it flows down into your contract: technical documentation, training data, logging, risk assessment.
- Selling a SaaS product with AI features to users in Europe. You may be the provider in the meaning of the Act.
- Embedding AI in recruitment or scoring features. That group sits in Annex III and receives the closest scrutiny.
Preparation that does not depend on the date
Whether your deadline is 2027 or 2028, these four items should start early because they accumulate over time and cannot be produced in a hurry:
- Record data provenance. Where training and fine-tuning data came from, and under what licence. This is the hardest thing to reconstruct after the fact.
- Operational logging. The system must record enough to reconstruct a decision: inputs, model version, output, and any human intervention.
- A human in the loop. For decisions affecting real people — hiring, credit — there must be a point where a person reviews, and that point must leave a trace.
- Marking AI-generated content. This is the 2026 date, and the cheapest item on the list.
One caution when reading news about the Act
There have been repeated rounds of lobbying and debate about postponing deadlines, so summary articles often contradict each other on dates. The safe approach: check the official reference timeline before putting a date into a plan, and note in your internal document which date the plan was based on and when it was checked.
For most Vietnamese companies, the practical value of reading this Act is not immediate compliance, but understanding the expectations European clients will bring into contracts signed this year.
Sources
- EU Artificial Intelligence Act — Implementation Timeline
- EU Artificial Intelligence Act — High-level summary
Dates checked on 24 September 2026 against artificialintelligenceact.eu. This article is not legal advice.
