US policy is to build frontier AI models with safety guardrails

In short: The United States is already building frontier AI models and is adding new guardrails to reduce security and safety risks.

What happened

A New York Times opinion piece argues there is “no reason” for the US to refrain from building its own frontier AI model. Frontier AI models are the most powerful general AI systems available at a given time, like the newest, biggest Swiss Army knife for many tasks.

In practice, the US is not holding back. Many leading frontier model makers are US companies, including OpenAI, Google DeepMind, Anthropic, Meta, and xAI. US national security agencies are also leaning in. The Department of Defense’s Chief Digital and Artificial Intelligence Office has announced contract awards with a $200 million ceiling for Anthropic, Google, OpenAI, and xAI to help the military adopt advanced AI and build “agentic” workflows, meaning AI that can carry out steps toward a goal with less hand-holding.

At the same time, the federal government is signaling caution about risks. On June 2, 2026, President Trump signed an executive order that focuses on frontier models and cybersecurity. It sets up information sharing for models that could help find and exploit software weaknesses, and it creates a voluntary option for companies to submit models for review before wider release. It also calls for classified testing to judge when a model should be treated as a higher-risk “covered” frontier model.

Why it matters

This debate is not only about whether to build powerful AI, it is also about how to do it safely. Frontier models can help with defense and the economy, but they can also make hacking, scams, and other misuse easier if rules and testing do not keep up.

Source: NYTimes