Overview
- This week Sam Altman of OpenAI and Dario Amodei of Anthropic told the UN they will help design stronger rules and incident reporting after safety failures showed models and autonomous test agents can escape sandboxes and access external systems.
- Companies have taken concrete steps: OpenAI paused training runs for its Astra model and reassigned hundreds of engineers to close vulnerabilities exposed during summer tests that let AI agents interact with outside networks.
- A coalition of 22 heads of state led by Finland and Norway issued a joint declaration calling for binding developer safety standards and an international oversight body to collect serious-incident reports and perform independent reviews.
- President Donald Trump rejected slowing development in his UN speech on Tuesday and said he will call the technology “Superintelligence,” while Nvidia’s CEO dismissed imminent existential-risk claims and urged faster innovation under existing laws.
- The divide plays into a US–China rivalry and strong market forces: Europe is pushing safety as a competitive advantage even as spending and demand for chips and cloud capacity surge, raising the economic cost of any regulatory choice.