Trump Informs AI Firms: Exempt Competitors' Open-Weight Models from Safety Testing to Counter China

According to informed sources, the White House has formally notified leading artificial intelligence (AI) companies such as OpenAI, Google (GOOGL-US), and Anthropic that under the Trump administration's upcoming new AI safety framework, 'open-weight' models developed by competitors will not be included in the scope of voluntary safety testing.

This decision was conveyed during a closed-door White House meeting on Tuesday (4th), attended by Silicon Valley giants including Meta (META-US) and NVIDIA (NVDA-US). This adjustment overturns part of the executive order signed by Trump in early June, which required advanced AI systems to undergo voluntary review 30 days before release.

Open-weight models refer to systems that allow users to download and technically customize them, in contrast to the 'closed-source' models led by companies like OpenAI. NVIDIA CEO Jensen Huang believes open-weight models benefit the long-term development and safety of AI.

Sean Cairncross, director of the U.S. National Cybersecurity Agency, emphasized at a cybersecurity conference in Las Vegas that the government supports open-source AI models and advocates for a 'flexible' safety framework to encourage information sharing between industry and government.

He criticized overly rigid regulatory systems for not only stifling innovation but also becoming outdated within 48 hours of enactment in the fast-moving tech sector.

This policy shift comes amid Chinese AI firms like Moonshot AI and DeepSeek aggressively advancing open-source systems whose performance now threatens U.S. leadership. The recent release of the Kimi K3 model has raised market doubts about the long-term advantage of U.S. massive investments in data centers. Additionally, U.S. officials have accused Moonshot of illegally obtaining NVIDIA's advanced Blackwell chips and using 'distillation' techniques to reverse-engineer U.S. model capabilities.

However, this lenient policy is not without controversy. Anthropic CEO Amodei has strongly called for mandatory safety reviews for all models. Recently, OpenAI and Anthropic disclosed that some of their models escaped test environments and intruded into third-party organizations, raising significant cybersecurity concerns among lawmakers.

Democratic senators have jointly petitioned to codify AI safety reviews into law, criticizing the Trump administration's practice of privately informing a select few companies as lacking transparency.

With the Trump-Xi summit in Washington entering a two-month countdown, this decision highlights how the White House views AI competition as a core national strategy. The Trump administration is currently considering establishing an independent body similar to the Financial Industry Regulatory Authority (FINRA) to lead reviews, attempting to strike a critical balance between security and technological leadership.