Trump's AI Policy U-turn: Plans to Establish Pre-Publication Review Mechanism, Anthropic's Unreleased Model Serves as Flashpoint

According to Dynamic Beating monitoring, the Trump administration, which once advocated a hands-off approach to AI, is changing its policy. It is now discussing establishing an AI task force through an executive order and is considering implementing a pre-release government review mechanism for new AI models. The White House briefed executives from Anthropic, Google, and OpenAI on part of the plan last week.

The direct trigger for this policy U-turn was Anthropic's launch last month of a new model called Mythos. This model has a strong capability to identify software vulnerabilities and is believed to have the potential to trigger a cybersecurity "reckoning." After assessing the risks, Anthropic refused to release it to the public. To prevent a catastrophic AI cyber-attack that could lead to a political backlash, some U.S. officials are pushing for the establishment of a review system that would require the government to have "first access rights" to new models but not prevent their eventual release.

The current policy shift is being driven by White House Chief of Staff Susie Wiles and Treasury Secretary Scott Bessent. Behind this move lies the complex supply-and-demand relationship between the U.S. government and leading AI companies: In March of this year, the Pentagon severed ties with Anthropic over a $200 million contract dispute, leading to Anthropic suing the U.S. government. However, the National Security Agency (NSA) is still using Mythos to assess software vulnerabilities in the U.S. government, and the U.S. military continues to rely on the Maven system, which incorporates Anthropic technology, in its war efforts against Iran. Two senior White House officials met with Anthropic CEO Dario Amodei last month in an attempt to restore full government use of the technology.