AI news
1. UN Ratifies "Global AI Safety Treaty" (GAIST) to Regulate Frontier Models
Date: August 7, 2026
The United Nations has officially ratified the Global AI Safety Treaty (GAIST), marking the first legally binding international agreement on artificial intelligence. Signed by 140 nations, the treaty mandates that any model exceeding $10^{26}$ FLOPs of training compute must undergo independent safety audits and incorporate hardware--level "kill--switch" protocols. This framework aims to standardize global safety benchmarks and prevent a "race to the bottom" in AI safety standards.
| Provision | Detail |
|---|---|
| Compute Threshold | > $10^{26}$ FLOPs |
| Mandatory Audit | Pre--deployment third--party verification |
| Enforcement | International AI Safety Council (IASC) |
Key Facts & Trends:
- Global Alignment: The treaty bridges regulatory gaps between the EU AI Act and US Executive Orders, creating a unified compliance landscape for multinational tech firms.
- Business Impact: Companies like OpenAI, Meta, and Google must now provide "Model Transparency Reports" to the newly formed IASC.
- Expert Commentary: "GAIST is the 'Montreal Protocol' for the AI era, ensuring that the pursuit of AGI does not compromise global security," says the UN Secretary--General.