Breaking

CERN detects signal beyond the Standard Model Mars Habitat Alpha: all 6 crew exit healthy after 72-hour test Quantum processor hits room-temperature coherence milestone Solar efficiency record broken — 47.6% under standard conditions Geneva AI Accord signed by 78 nations CRISPR 3.0 achieves 99.3% accuracy in mitochondrial editing CERN detects signal beyond the Standard Model Mars Habitat Alpha: all 6 crew exit healthy after 72-hour test Quantum processor hits room-temperature coherence milestone Solar efficiency record broken — 47.6% under standard conditions Geneva AI Accord signed by 78 nations CRISPR 3.0 achieves 99.3% accuracy in mitochondrial editing

← Back

✍️ Opinion

The Geneva Accord Is Well-Intentioned — But It Misses the Most Dangerous Systems

Opinion: The treaty covers visible AI systems while leaving military, intelligence, and infrastructure applications entirely outside its scope.

Elena Kovacs, AI Policy Editor March 7, 2026 5 min read

OPINION

Article 7, subsection 4(b) of the treaty exempts AI systems used for "national defence, military operations, or intelligence activities" from all IAMA oversight requirements. This single carve-out excludes the most dangerous AI systems in existence today.

"We have regulated the shop window while leaving the back room entirely dark."

— Elena Kovacs

The accord needed a risk-tiered approach that placed the highest burdens on the highest-stakes systems. We should be honest: what we have regulated is the visible part of a very large iceberg.

The Geneva Accord Is Well-Intentioned — But It Misses the Most Dangerous Systems

Related Articles

78 Nations Sign History's First Binding Global AI Law — And Scientists…

When the Machine Designs the Experiment: Inside the World's Most Advance…

CERN's New Signal: The Discovery of the Century, or a Beautiful Statisti…