We're encouraged by the White House’s Executive Order directing federal agencies to evaluate frontier AI models and establish a framework for government oversight before these models are released to the public.

Oversight of AI is an increasingly urgent issue. As we saw earlier this year with Anthropic’s Claude Mythos, AI companies are developing systems that present significant threats to cybersecurity and more — and they’re building these systems largely in secrecy, with no independent check on their capabilities or risks.

AI companies cannot be left to grade their own homework on these matters. When AI systems are powerful enough to impact national security, critical infrastructure, and our everyday lives, the public has a right to know these systems have been deemed safe through independent assessments. This Executive Order is a much-needed incremental step toward greater transparency and accountability for the AI industry. And it signals that the government can take quick action when faced with this rapidly evolving tech.

The administration has emphasized in this Executive Order that it wants to support the deployment of the “best and most secure technology”; delivering on that goal will require clear standards for assessing frontier models and concrete measures to protect all Americans from unsafe AI products.

As we outlined in the AI Roadmap, getting to a better future with AI will require a whole-of-society approach. National security assessments of frontier models are one critical component of a set of necessary actions. We look forward to seeing this Executive Order implemented in the coming months and to this administration and Congress continuing to build upon its foundation.

Read our AI Roadmap here: humanetech.com/ai-roadm…

Read more about the new EO: washingtonpost.com/tech…