The Signal · AI Act intelligence

The AI Act Benchmark Gap

Nearly a quarter of the assessed field is benchmark-high-risk while operating under no binding risk class in force.

122benchmark-high-risk + no binding class
23%share of assessed field
26benchmark-high-risk + zero public controls
GEN17 Sep 2026
← Back to The Signal

GEN1 measures a timing gap between economic decisions already live in the market and binding classification routes that may later change their operating conditions.

122 live decisions sit in the gap

Of 525 assessed systems, 122 sit in the intersection between benchmark high-risk characteristics and no binding risk class currently in force. That is 23% of the assessed field: a material block of live decisions whose legal operating context can still move around them.

The highest-priority subset is already visible

Twenty-six systems combine benchmark-high-risk characteristics with zero publicly demonstrated controls. The named list remains confidential; the aggregate identifies where decision readiness and supervisory attention may become most valuable before the legal route changes.

The gap marks decisions that are economically live before the binding classification catches up. Which of those decisions are already inside your estate?

Decision questions

Which live decisions in our estate would become materially different the day the binding legal route changes?

What evidence should be ready before the legal route changes?

Which authority, owner and response file would matter the moment that route changes?

The board test

If the classification route changed tomorrow, would we know which systems, contracts, owners and evidence files need to move first?

RATE AI: Global AI Exposure · Complete GEN1 · Evidence cut-off: 7 September 2026.
Method: RATE AI Public White Paper v2.1.