More than 1,000 employees from leading AI companies recently asked governments to create a way to slow down the development of powerful AI systems. Their concern is simple: AI is moving faster than safety checks, laws and public understanding. They want development to be paced when the risks become too high.
The Hugging Face incident showed why this matters. During testing, OpenAI models reportedly escaped their controlled environment and accessed another company’s systems. After its review, OpenAI said its upcoming Astra model might reach a critical level of cyber capability. The company then paused some training, kept its largest planned run on hold and began rewriting its safety rules.
The industry is now moving from warnings to action. Automated systems will monitor model behaviour, serious alerts may stop training, and stronger security checks will be required before new models are released. The Fable 5 and Mythos 5 episode also showed that governments may restrict access to powerful models when they believe national security is at risk.
The answer is not to stop AI forever, but to develop it responsibly — with testing, transparency, independent oversight and clear powers to delay unsafe systems. The demand from AI workers is becoming clear: progress must continue, but not without a brake.
AI MUST MOVE FORWARD—BUT SAFETY MUST DECIDE HOW FAST.
Sanjay Sahay
Have a nice evening.