Vietnam has moved to bar the illegal use of state secrets and defense or security data to train artificial intelligence models, tightening control over one of the fastest-growing and most sensitive uses of data as governments race to regulate AI.
Vietnam bans AI training on state secrets

The measure is part of the country’s newly passed Data Security Law, approved by 468 of 474 lawmakers and set to take effect on July 1, 2027. It gives Hanoi a broader legal framework for data classification, audits and incident response, while drawing a hard line around information that touches national security, public order and state confidentiality.
The law matters economically because it raises the compliance threshold for companies handling data in Vietnam, particularly telecom operators, cloud providers, data centers and AI developers. It also signals that governments are increasingly treating training data as a strategic asset rather than a passive input, with implications for cross-border platforms that process Vietnamese data or operate models trained on local datasets.
For investors, the new rules add a regulatory risk premium to AI infrastructure and software exposure in Vietnam and across the region. Companies that rely on broad data ingestion to improve models may face tighter access to sensitive datasets, more audits and potential delays if authorities require approval for using higher-risk defense and security data. The law also explicitly targets “data poisoning,” or the insertion of false or manipulated data into training sets, reflecting official concern that AI systems can be compromised at the model level, not just through cyber intrusions.
The legislation applies not only to Vietnamese entities but also to foreign organizations and individuals in Vietnam, and in some cases to those outside the country when they process data under Vietnam’s jurisdiction or affect its data security. That extraterritorial reach increases the likelihood that global cloud and AI firms will need local compliance frameworks rather than relying on broad international policies.
Vietnam’s approach also shows how AI regulation is moving beyond privacy and toward national-security enforcement. Data are to be divided into five risk levels, with stricter impact assessments and audits for level 3 and above, and mandatory reporting for level 4 and 5 data. Authorities said protection must begin at the design stage and continue through the entire data lifecycle, from collection and storage to transmission, deletion and destruction.
The bull case for the sector is that clearer rules can support trust and accelerate institutional adoption of digital services if companies can demonstrate stronger security controls. The bear case is that tighter limits on training data, combined with audit and incident-reporting obligations, could slow product development and raise operating costs, especially for firms that depend on rapid model iteration.
For markets, the immediate impact is unlikely to be on listed U.S. megacap names alone, but the wider message is unmistakable: AI growth is now colliding with sovereign data rules. That is likely to favor vendors with strong governance, localized hosting and compliance tooling, while pressuring operators whose business models depend on frictionless access to data at scale.
The law’s 2027 start date gives companies time to adjust, but it also sets up a two-year runway for tougher procurement standards, contract revisions and possible restrictions on where sensitive data can be stored or processed. In an AI market still defined by the search for more compute and more data, Vietnam has chosen to make security the binding constraint.
| Entity | Gains | Losses |
|---|---|---|
| Vietnamese state | ▲Stronger data control | ▼Less model openness |
| Cloud and telecom providers with compliance tools | ▲Higher demand for security services | ▼Higher audit burden |
| AI developers using broad training data | ▲Clearer legal framework | ▼Restricted access to sensitive data |
| Foreign platforms processing Vietnamese data | ▲Regulatory clarity | ▼Extraterritorial compliance risk |


