Regional summit agrees on a common licensing framework for high-risk AI models

Six countries closed the regional AI governance summit by agreeing on a common licensing framework for high-risk models — a first for the region — requiring developers of large models to register with a regulator and pass an independent audit before commercial release.

What the framework says

Systems are sorted into three risk tiers. The top tier covers hiring, credit, medical diagnosis and critical-infrastructure management; it demands full documentation of training data and evaluation methods, plus a clear mechanism for human intervention in the final decision.

The middle tier covers recommendation systems, voice assistants and content-generation tools, which must disclose their use of AI and label generated content. The lowest tier stays outside prior licensing altogether.

A transition period and graded penalties

Signatories granted companies eighteen months to comply. Penalties start at fines of up to three percent of annual revenue and can escalate to service suspension for repeat offences.

"We do not want to slow innovation; we want to know who is accountable when a system gets it wrong," the head of the regulatory committee said at the closing press conference.

How the industry reacted

Start-ups welcomed the clarity, while larger companies raised concerns about the cost of independent audits. The experts MADAR spoke to disagreed on three points:

• Whether new regulators can attract the technical expertise audits require.
• How to treat open-source models with no single commercial entity behind them.
• Aligning the framework with European legislation to avoid double compliance.

  

Closing session of the regional AI governance summit

What happens next

Implementing regulations are expected within six months, with the first registrations opening early next year. MADAR is following the framework and its impact on users in a dedicated series.