CAIDP's Proposal for Vietnam to Join the AI Treaty and the Legal Technical Differences Between 'Signature' and 'Accession' / Reflections from the Perspective of 2026

0 Introduction

On January 30, 2026, the Center for AI and Digital Policy (hereinafter referred to as "CAIDP") published a document urging the Socialist Republic of Vietnam to apply for observer status at the Council of Europe's Committee on Artificial Intelligence (CAI) and to sign and ratify the Framework Convention on Artificial Intelligence and Human Rights, Democracy and the Rule of Law (hereinafter referred to as the "AI Treaty").

The CAIDP document acknowledges that Vietnam is already accelerating its domestic AI governance, while pushing it to connect to an international, legally binding framework.

This type of proposal is becoming a familiar sight in the context of AI governance. However, in the world of international treaties, "signature," "ratification," and "accession" are similar but distinct acts, and the landscape of the discussion changes based solely on the entry provisions (eligibility for signature or requirements for accession). As the debate over AI regulation intensifies, such mundane provisions are relatively easy to overlook. Therefore, this article uses the CAIDP proposal as material to organize, in order, the relationship between CAI observer participation and the AI Treaty, the status of Vietnam's domestic legal development, the legal technical differences between "signature" and "accession," and the implications for corporate practice.

1 The Location of the Problem

AI governance is not completed by domestic law alone. As models, data, and supply chains cross borders, countries build systems with an awareness of interoperability. The extraterritorial reach of the EU AI Act is one example, and Vietnam's adoption of risk-based classification and transparency obligations is not unrelated to international trends.

On the other hand, from the perspective of international law, the AI Treaty is positioned as the first legally binding general framework in the context of the Council of Europe and can serve as an external yardstick to encourage the development of domestic laws in each country.

However, a treaty is not a universal switch. The effectiveness of a treaty only gains substance through the processes of signature, ratification (or approval), entry into force, and domestic implementation. Furthermore, whether a non-member state can "sign" is determined by the final provisions of the treaty text. Here, a legal technical question arises in interpreting the CAIDP proposal.

2 The Framework of the CAIDP Proposal

The demands of the CAIDP document are clear. First, apply for observer status at the CAI. Second, sign and ratify the AI Treaty. In addition, CAIDP lists Vietnam's enactment of AI laws and the development of digital-related legislation (personal data protection, cybersecurity, etc.), emphasizing that the domestic institutional foundation is being prepared.

The logic of CAIDP is that Vietnam has already embodied many of the requirements sought by the AI Treaty in its domestic law. Therefore, it argues that joining the treaty is the best means to position these domestic reforms within an international legal framework and to guarantee their legitimacy and irreversibility.

3 "Signature" or "Accession": Misreading the Entry Provisions

However, if we shift gears here and look at the text of the treaty, the map changes.

Article 30 of the AI Treaty (CETS No. 225) defines the eligibility for signature. According to paragraph 1 of the same article, this treaty is open for signature by member states of the Council of Europe, non-member states which have participated in its elaboration, and the European Union.

The non-member states that participated in the elaboration of the treaty are the 11 countries of Argentina, Australia, Canada, Costa Rica, the Holy See, Israel, Japan, Mexico, Peru, the United States, and Uruguay. Vietnam is not included among them.

On the other hand, Article 31 provides for "accession" as a path for non-member states that did not participate in the elaboration to join. According to paragraph 1 of the same article, after the entry into force of this treaty, the Committee of Ministers of the Council of Europe may, after consultation with the parties to the treaty and obtaining their unanimous consent, invite non-member states that did not participate in the elaboration to accede.

In other words, since Vietnam is not a participating state in the elaboration, the scenario of immediately "signing" is not natural based on the treaty wording, and it is more consistent to frame it as a discussion of "accession."

This point is not merely a matter of terminology. Signature is a procedure granted to entities that meet the entry qualifications under the treaty, while accession is a mechanism for existing parties to welcome an external country after the treaty enters into force. Since accession is an invitation-based system and requires the unanimous consent of existing parties, the context of politics and diplomacy (the evaluation, trust, and mutual interests of the parties) is strongly reflected.

It makes more sense to understand CAIDP's statement that "observer participation is a procedural prerequisite" not as a strict legal requirement, but as a political and institutional stepping stone to obtain an invitation. It is a strategy to pave the way for a future invitation to accede by accumulating a track record at the CAI and building relationships of trust with the parties.

4 Development Status of Vietnam's Domestic Laws: Adaptation of the EU Model and Local Adjustments

Vietnam developed relevant legislation in rapid succession between 2025 and 2026.

First, the Law on AI (No. 134/2025/QH15), passed on December 10, 2025, will come into effect on March 1, 2026. Strongly referencing the EU AI Act, this law adopts a framework that classifies AI systems into high-risk, medium-risk, and low-risk categories, imposing obligations on both providers and deployers regarding documentation, notification, transparency, and the labeling of AI-generated content. A distinctive feature is the inclusion of a category for systems that 'can mislead and influence users in a way that is not easily recognizable as AI,' thereby bringing generative AI and deepfakes within the scope of the regulation.

Next, the Vietnam Personal Data Protection Law (No. 91/2025/QH15) was passed on June 26, 2025, and came into effect on January 1, 2026. Heavily influenced by the GDPR, this law establishes penalty provisions for serious violations and a 72-hour notification requirement in the event of a data breach. Of note from an AI governance perspective is the provision regarding data protection in big data processing and artificial intelligence (Article 30), which mandates the integration of appropriate data security measures in AI systems.

Furthermore, the Vietnam Cybersecurity Law (No. 116/2025/QH15) was passed on December 10, 2025, and will come into effect on July 1, 2026.

When viewed together, it can be said that Vietnam is transitioning toward a legal system that bundles risk classification, transparency, data protection, and cybersecurity, with the enactment of the AI Law at its apex. What the AI Treaty also seeks is precisely the establishment of risk management and remedies from the perspective of human rights, democracy, and the rule of law throughout the entire lifecycle of AI systems.

5 Layering Soft Law and the Hardening Agent Called a 'Treaty'

International AI governance does not progress through treaties alone. UNESCO's Recommendation on the Ethics of AI and its implementation support tool, the Readiness Assessment Methodology (RAM), diagnose the systems and capabilities of each country and present policy menus. Regarding Vietnam, UNESCO published a RAM report in 2025.

The OECD AI Principles (OECD/LEGAL/0449) have also influenced the institutional design of various countries as an international standard that, while lacking strong binding force, is highly referential. At the regional level, policy coordination within ASEAN has also progressed, and the Hanoi Digital Declaration was adopted at the 6th ASEAN Digital Ministers Meeting held in Vietnam in January 2026.

Thus, a multi-layered structure is clearly visible: treaties as hard law, RAM and principles as soft law, and regional coordination as an intermediate layer. A treaty does not merely set forth principles; it requires member states to take domestic measures and generates peer-review-like pressure through reporting on implementation status and the Conference of the Parties. There can be a sequence where the contours of principles expanded by soft law are bonded to legal procedures by a treaty. CAIDP's proposal likely views the AI Treaty as this bonding point.

6 Confirmation of Numbers and Legal Status

The handling of numbers surrounding the treaty is also an area that should be verified with primary sources. While the CAIDP document states that '45 countries have signed,' caution is required regarding this figure.

At the time of the treaty's adoption on May 17, 2024, those participating in the formulation were 46 Council of Europe member states, the EU, and 11 non-member states. However, these were participants in the formulation, not signatories. When the treaty was opened for signature in Vilnius on September 5, 2024, the 10 entities that actually signed were Andorra, Georgia, Iceland, Norway, the Republic of Moldova, San Marino, the United Kingdom, Israel, the United States, and the EU.

The treaty enters into force upon ratification by five countries (including at least three Council of Europe member states). It was considered to have met the requirements for entry into force on November 1, 2025, and the United Kingdom, France, Norway, and others have completed ratification. Confusion between the concepts of 'signing,' 'endorsing,' and 'party status' easily creeps in here. Since AI governance discussions are often swayed by political catchphrases, one should always verify numbers and legal status with primary sources.

7 Implications for Corporate Practice

From a practical perspective, what is important is that regulation is not monolithic but layered. The Vietnam AI Law centers on risk classification, transparency, and labeling, clarifying the division of roles between providers and users. The Personal Data Protection Law and the Cybersecurity Law cross-sectionally tighten the governance of data handling and the information space.

If the AI Treaty (becoming a party through accession) is added to this in the future, vertical requirements such as human rights impact assessments and access to remedies derived from the treaty could become an opportunity to reorganize domestic legislation.

Therefore, in practice, it is necessary to manage at least the following three points separately. First, the classification of the AI system itself (high, medium, or low risk) and the corresponding technical documentation, notifications, labeling, and supervisory responses. Second, data security controls, including the processing and cross-border transfer of personal data, log management, and incident response. Third, the transparency (labeling, etc.) of content generation and editing, including generative AI and deepfakes, and the reduction of risks related to misidentification and fraud.

When the possibility of treaty accession becomes an issue, political and economic factors regarding what the party state will invite also enter the picture. While companies cannot directly control this, it should not be overlooked that their own compliance posture (transparency, explainability, cooperation with supervision, and design of remedies) serves as material for building trust with counterpart authorities and business partners.

8 Conclusion

AI governance is driven more by mundane articles than by flashy slogans. While provisions regarding signature eligibility, accession requirements, and entry-into-force conditions are often placed at the end of a treaty, they are, in fact, the articles that should be read first.

While the CAIDP's proposal outlines a broad direction for connecting Vietnam's institutional development to international frameworks, it also serves as a reminder of the legal technical weight carried by the term 'signature.' As discussions surrounding the AI treaty expand, it becomes increasingly necessary to return to primary sources more frequently. AI crosses borders, but laws do not cross articles. Therein lies both the tediousness and the intrigue of the subject.

(Magazine) 'AI and Law - Reflections'

* Please refer to the following for the table of contents

Note General Terms of Service Article 3, Paragraph 2, First Sentence
3.2 The copyright of digital content produced by the creator belongs to the creator.