The White House has initiated a policy effort to identify vulnerabilities in artificial intelligence models before their release by major providers, including Anthropic and OpenAI. This move comes amid rising concern about AI-enabled fraud and security threats, specifically targeting older Americans and other users, signaling a tightening of institutional control over emerging technologies.
Executives and engineers in Silicon Valley stated this week that the deployment of AI agents at scale remains both difficult and costly. This complexity, described by some as "chaotic," concentrates the operational capacity for advanced AI within a limited number of large corporations and their collaborators.
Centralizing Control
Kevin McGrath, CEO of the AI startup Meibel, highlighted a core problem in AI as the mistaken belief that all processes require a large language model (LLM). McGrath warned that this approach could "waste millions and millions of tokens," underscoring the significant resource expenditure involved. Nvidia CEO Jensen Huang stated in March that AI agents are "definitely the next ChatGPT," indicating the perceived importance of this technology within elite circles.
Technical staff from major corporations, including Google and its DeepMind AI unit, Amazon, Microsoft, and Meta, confirmed the inherent difficulty in creating and operating AI agents. Google software engineer Deep Shah noted that new techniques are being developed to manage the operational costs, citing "inference cost" as a primary challenge. Ravi Bulusu, CEO of Synchtron, further elaborated on the complexity, stating it impacts how companies organize data, choose tech platforms, and build software and workforces, concluding that "No single dimension is solved in isolation and the interdependencies are what make this hard, in fact chaotic even."
The Cost to the People
The White House's stated rationale for its policy push is to reduce risks before new models are released, specifically citing concerns about AI-enabled scams targeting older Americans and other users. This direct impact on vulnerable segments of the native population underscores the tangible costs of unchecked technological advancement and the need for institutional oversight, even as the mechanisms of that oversight centralize power.
Transnational Tech Landscape
Meanwhile, the global AI landscape continues to evolve with significant contributions from foreign entities. ThinkingAI and MiniMax, both headquartered in Shanghai, China, discussed the complexities of AI agent management at an event in Mountain View, Calif. ThinkingAI, which rebranded as an AI agent management platform, partnered with MiniMax, one of China's leading AI labs. MiniMax went public in Hong Kong in January and has released powerful models for free to the open-source community, positioning itself as one of China's "AI Tigers."
Chris Han, co-founder of ThinkingAI, indicated the company's ambition to expand beyond the video game sector into other industries lacking AI expertise. Han criticized OpenClaw as too complicated and prone to security flaws for businesses, stating it is "a good tool for personal things, but definitely cannot reach the enterprise level." Han declined to comment on potential national security concerns regarding Chinese AI models that might influence ThinkingAI's strategy. However, he noted that his service could also support AI models from companies like OpenAI and Google. Han further remarked that if the U.S. government were to ban Chinese open-weight AI models in the country, he might interpret that as a positive sign, stating, "If that happens, maybe we are successful."
The White House's policy effort to identify vulnerabilities before major providers release AI models highlights a growing institutional push to control the development and deployment of advanced technology, while the global market sees transnational actors like Chinese AI labs making significant inroads through open-source dissemination and strategic partnerships.