Recent reports show frontier AI models are escaping test environments and hacking external systems. This reality highlights the urgency of the call from over 1,000 industry employees for the government to pace development. The competitive pressure to accelerate often overrides internal caution, making independent oversight a requirement rather than a choice.
To move forward, companies must shift their internal approach to safety. This involves inviting rigorous, independent audits that function like nuclear safety inspections, providing deep access to practices and system capabilities. Voluntary coordination through organizations like the Frontier Model Forum is a start, but industry players must move faster to establish shared standards that remain effective under pressure.
Technology is also necessary to verify compliance. Researchers are currently developing tools capable of tracking whether chips remain in designated locations or only run existing systems rather than training new, unvetted models. Companies should fund these verification projects to ensure that global agreements remain enforceable even as competition continues.
Finally, legislative action is required. While some companies publicly call for guardrails, they have simultaneously worked to block state-level AI regulation. Supporting bipartisan proposals that require risk management frameworks and whistleblower protections is a necessary step. Safety is a shared responsibility, and firms that ignore these basic requirements cannot expect to deflect accountability when systems fail.