This stance comes as prominent industry leaders from Anthropic, OpenAI, and xAI have suggested a slowdown in development while seeking narrow legal waivers or limited liability for AI-fueled harms.
The Treasury Department’s opposition to these "blank checks" on liability is rooted in recent security concerns, specifically those involving the Anthropic model Mythos and its associated cybersecurity risks.
By maintaining strict liability, the administration aims to prevent "regulatory capture"—a situation where large companies influence regulations to their own advantage—which Bessent warned would stifle innovation and competition.
This approach emphasizes that the financial sector and other critical infrastructure must remain resilient against evolving threats without granting legal immunity to the technology's developers.
To further balance safety with competitive goals, Bessent advocated for the domestic development of open-source AI models to counter technological advancements from China.
The Treasury continues to coordinate with major banks and the Cybersecurity and Infrastructure Security Agency through the "Gold Eagle" clearinghouse, a mechanism used to share vulnerability scans and software patches across sectors.
These efforts follow a July cyberattack on the platform Hugging Face, highlighting the administration's focus on integrating cybersecurity lessons from the financial industry into the broader technological landscape.